[
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b9d6069139dc3b8c68afcd88459395e754cf2eb76cd52529c50acbe84221915e",
    "id": "007dd7dea4f9c9ba8cead6f1a649",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.7081804275512695,
    "wer": 0.0,
    "identity": 0.5627717971801758,
    "voiceclap": -0.2463889718055725,
    "fitness": 0.777935930150462,
    "legacy_fitness": 0.777935930150462,
    "target_cdf": 0.8413447460685429,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2503652572631836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6307411411317138,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2503652572631836,
        "normalized": 0.8413447460685429,
        "contribution": 0.42067237303427146
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0873427391052246,
        "normalized": 0.2458838571357379,
        "contribution": 0.061470964283934476
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8853464126586914,
        "normalized": 0.607741561467731,
        "contribution": 0.09116123422015963
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6478986740112305,
        "normalized": 0.574365695933482,
        "contribution": 0.0574365695933482
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/00/007dd7dea4f9c9ba8cead6f1a649.mp3",
    "audio_sha256": "a816cbe72c291a904af2349410ac0589cadd23da074fb6c6a9ce0da7421b1b71",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e04f01ba133d2082f620ec32310bc1650add740e1fd3f493d5adf85dd8a6a610",
    "id": "00ee6020e3eb4889f9fb85c43de7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.483288764953613,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0007413662970066071,
    "fitness": 0.8200460838024388,
    "legacy_fitness": 0.8200460838024388,
    "target_cdf": 0.9696036431679269,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1593828201293945,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7374000778290046,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1593828201293945,
        "normalized": 0.9696036431679269,
        "contribution": 0.48480182158396345
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.466392755508423,
        "normalized": 0.8196579878328937,
        "contribution": 0.08196579878328937
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fcf255abe44b70c01cf79b5590d4731f9ed399c42a6bef040cdf8e2c8637e516",
    "id": "00fe8bcf6af8cfe89810b5666edf",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.9801597595214844,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04270058125257492,
    "fitness": 0.8163834048515671,
    "legacy_fitness": 0.8163834048515671,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7132635134596915,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4427849054336548,
        "normalized": 0.07640335680063054,
        "contribution": 0.019100839200157635
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.068478107452393,
        "normalized": 0.9053242349151492,
        "contribution": 0.13579863523727237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9053282737731934,
        "normalized": 0.6613007191610654,
        "contribution": 0.06613007191610654
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/00/00fe8bcf6af8cfe89810b5666edf.mp3",
    "audio_sha256": "6824969a23717bc4faba936acdaed887b92dd08be12125942db0bdd0eaa621cb",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9107de5fc27fd97a57c1b60b4c74a516411e2cc14415e77742703f4b80bd0972",
    "id": "016c2e9cc571f3979cb15c227828",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09883516281843185,
    "fitness": 0.7842633812176121,
    "legacy_fitness": 0.7842633812176121,
    "target_cdf": 0.8855360207054879,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4613871574401855,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.653654558125175,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4613871574401855,
        "normalized": 0.8855360207054879,
        "contribution": 0.44276801035274393
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9245071411132812,
        "normalized": 0.1907869738584032,
        "contribution": 0.0476967434646008
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.112187385559082,
        "normalized": 0.6817707678889575,
        "contribution": 0.10226561518334362
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.749110221862793,
        "normalized": 0.6092418912448665,
        "contribution": 0.060924189124486654
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "28f1b9cd68a30a6b3f5b9c9f728feb4ea724ea762f539c01320267e42438d1cb",
    "id": "016f861cbc0fa4c6706f0f5bc3a2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.8546390533447266,
    "wer": 0.0,
    "identity": 0.5866890549659729,
    "voiceclap": -0.03985682874917984,
    "fitness": 0.8180018306359391,
    "legacy_fitness": 0.8180018306359391,
    "target_cdf": 0.9993562461107455,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.555373668670654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7871738859853025,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.555373668670654,
        "normalized": 0.9993562461107455,
        "contribution": 0.49967812305537274
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9211146831512451,
        "normalized": 0.1897260605487444,
        "contribution": 0.0474315151371861
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.90467643737793,
        "normalized": 0.9796648065278896,
        "contribution": 0.14694972097918343
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.1088666915893555,
        "normalized": 0.9311452681356018,
        "contribution": 0.0931145268135602
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4b1e12a9c83bc9808ed1e7e549be8685182bf8871db1acd09e92c1ddd5670855",
    "id": "0184c8914d0f75d29c68c76c62fd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.02966121956706047,
    "fitness": 0.8208691810555948,
    "legacy_fitness": 0.8208691810555948,
    "target_cdf": 0.9690622995272662,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.151266574859619,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7666349484642919,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.151266574859619,
        "normalized": 0.9690622995272662,
        "contribution": 0.4845311497636331
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2060770988464355,
        "normalized": 0.29089184034451254,
        "contribution": 0.07272296008612814
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7037534713745117,
        "normalized": 0.8394469854437661,
        "contribution": 0.1259170478165649
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.532400369644165,
        "normalized": 0.8346379079796576,
        "contribution": 0.08346379079796577
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e52dc78037fcd1c9149d18d0a010642d4fc79ddddbe75c19095590f76c3c0f13",
    "id": "018b97e0d58ed3fc9296d7cd1c6c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": 0.6236056089401245,
    "voiceclap": -0.016281990334391594,
    "fitness": 0.7705669915503611,
    "legacy_fitness": 0.7705669915503611,
    "target_cdf": 0.8985686055474182,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.534433364868164,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7152020464392548,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.534433364868164,
        "normalized": 0.8985686055474182,
        "contribution": 0.4492843027737091
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.063595771789551,
        "normalized": 0.23735276923124937,
        "contribution": 0.05933819230781234
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.685961961746216,
        "normalized": 0.835607112993103,
        "contribution": 0.12534106694896543
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.43558931350708,
        "normalized": 0.8123848440876791,
        "contribution": 0.08123848440876791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "71a0602cea4f713ba8a55ad2c5e87c02ccb8bcb106312f48980cfa365fde0d2e",
    "id": "01cfb8971629062d3e6f31a6c30a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.570657730102539,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.042648110538721085,
    "fitness": 0.8023238676196279,
    "legacy_fitness": 0.8023238676196279,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7194547129442811,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0839502811431885,
        "normalized": 0.24465512789027288,
        "contribution": 0.06116378197256822
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.605900287628174,
        "normalized": 0.8175982108486767,
        "contribution": 0.12263973162730149
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.175959348678589,
        "normalized": 0.7441683469516106,
        "contribution": 0.07441683469516107
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c2fd29c8be12fea579e66bf40bfbb28b7de68dedf969a3312cd02ac2232f2e34",
    "id": "0225d332a6739f736c7c04453a36",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.074300289154053,
    "wer": 0.0,
    "identity": 0.7678083181381226,
    "voiceclap": 0.020458102226257324,
    "fitness": 0.8258796552612797,
    "legacy_fitness": 0.8258796552612797,
    "target_cdf": 0.9995867291677825,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.6852335929870605,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7817159000486208,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.6852335929870605,
        "normalized": 0.9995867291677825,
        "contribution": 0.49979336458389123
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8702285289764404,
        "normalized": 0.17425071298327421,
        "contribution": 0.043562678245818554
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.833510875701904,
        "normalized": 0.9763929720149331,
        "contribution": 0.14645894580223998
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.012055397033691,
        "normalized": 0.91900911416671,
        "contribution": 0.091900911416671
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d280ed99f6c134f6d090574d44c432ea2b6b502d4929df71703e4d505eb3b016",
    "id": "023f9306ef2c503abb63997197ba",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": 0.7043393850326538,
    "voiceclap": -0.0017036087810993195,
    "fitness": 0.7807142004011786,
    "legacy_fitness": 0.7807142004011786,
    "target_cdf": 0.9154342620847058,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.639944076538086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7091870534266391,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.639944076538086,
        "normalized": 0.9154342620847058,
        "contribution": 0.4577171310423529
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8905830383300781,
        "normalized": 0.18034202015435835,
        "contribution": 0.04508550503858959
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.730440616607666,
        "normalized": 0.845096013305189,
        "contribution": 0.12676440199577835
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.369581699371338,
        "normalized": 0.7962001534991823,
        "contribution": 0.07962001534991824
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "abf8ca682f75cf1af41230bac3bfd14f8087893a31a72c4bbbeb86df1e95d69b",
    "id": "024c6174b43e3e9dedf64a19a4dc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.055472373962402,
    "wer": 0.0,
    "identity": 0.6813439130783081,
    "voiceclap": -0.08129168301820755,
    "fitness": 0.7388665056181359,
    "legacy_fitness": 0.7388665056181359,
    "target_cdf": 0.8175982348029629,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.152970552444458,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5900828888218056,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.152970552444458,
        "normalized": 0.8175982348029629,
        "contribution": 0.40879911740148145
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.598835825920105,
        "normalized": 0.10564978306864803,
        "contribution": 0.026412445767162007
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1522183418273926,
        "normalized": 0.6942076233845773,
        "contribution": 0.1041311435076866
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4575767517089844,
        "normalized": 0.5074018214547558,
        "contribution": 0.05074018214547558
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b444d08f684973e1a2e14cfeb2eae44092c6a06266be325a01d9428cd5803c16",
    "id": "0305b5416db5cae2d23a6441fbf2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.30960172414779663,
    "fitness": 0.8203997520142521,
    "legacy_fitness": 0.8203997520142521,
    "target_cdf": 0.9638222590155852,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0782203674316406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7032308809282887,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0782203674316406,
        "normalized": 0.9638222590155852,
        "contribution": 0.4819111295077926
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.494703769683838,
        "normalized": 0.7906255737437923,
        "contribution": 0.11859383606156884
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9647350311279297,
        "normalized": 0.6803758083890165,
        "contribution": 0.06803758083890164
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/03/0305b5416db5cae2d23a6441fbf2.mp3",
    "audio_sha256": "c520a1e6dbf967636dd675d1b4c78a5977b3b3aec4eaad602f4d04672084c38e",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1fed6e5378e978e2660f310bbebfd2f1720cf47ccfb8628042ec017509956990",
    "id": "03103b9ac9342dff912a05520a8b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.18099308013916,
    "wer": 0.03571428571428571,
    "identity": 0.4213966131210327,
    "voiceclap": -0.042089931666851044,
    "fitness": 0.7956820436418429,
    "legacy_fitness": 0.7956820436418429,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7330464566247672,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8816680908203125,
        "normalized": 0.8746060247794507,
        "contribution": 0.1311909037169176
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2639694213867188,
        "normalized": 0.7686398657383282,
        "contribution": 0.07686398657383282
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ef99b3343e9e2ba6dfd8f1e2c03e0bba052393b3d24e13c6c4cd3b9ac583b531",
    "id": "03115cc06a8efd1704afe4bde551",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.808597087860107,
    "wer": 0.06666666666666667,
    "identity": 0.23140448331832886,
    "voiceclap": -0.08361271023750305,
    "fitness": 0.7500337570734175,
    "legacy_fitness": 0.7500337570734175,
    "target_cdf": 0.8914684095185468,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.493852138519287,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.627197902068623,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.493852138519287,
        "normalized": 0.8914684095185468,
        "contribution": 0.4457342047592734
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.84308922290802,
        "normalized": 0.16633496146433524,
        "contribution": 0.04158374036608381
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.920929193496704,
        "normalized": 0.6196978959173229,
        "contribution": 0.09295468438759844
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.349764347076416,
        "normalized": 0.4692527255566742,
        "contribution": 0.04692527255566742
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5158c0629bbf8d7a1ef23ad1c9cb7329f33e12a924ad0cb271f0790bd58cf42f",
    "id": "0328767437fd324f488763dfba17",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": 0.6073934435844421,
    "voiceclap": 0.005025139078497887,
    "fitness": 0.8375878842503635,
    "legacy_fitness": 0.8375878842503635,
    "target_cdf": 0.9935110473296658,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7924485206604004,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7840948739145159,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7924485206604004,
        "normalized": 0.9935110473296658,
        "contribution": 0.4967555236648329
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1992921829223633,
        "normalized": 0.28821948574397915,
        "contribution": 0.07205487143599479
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.015103816986084,
        "normalized": 0.8971763038367273,
        "contribution": 0.1345764455755091
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4135866165161133,
        "normalized": 0.807080332381791,
        "contribution": 0.0807080332381791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "acd96f42a8db13985b22117f9c6de0c77417d1aa3ee06889ee2077c397b8e11b",
    "id": "03500fed9d9db070b68d2edf39a6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.5786004066467285,
    "genuineness": 4.2249250411987305,
    "wer": 0.0,
    "identity": 0.5445773601531982,
    "voiceclap": -0.038390226662158966,
    "fitness": 0.7630515588041745,
    "legacy_fitness": 0.7630515588041745,
    "target_cdf": 0.8450960105094518,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2665977478027344,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5964682495060557,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2665977478027344,
        "normalized": 0.8450960105094518,
        "contribution": 0.4225480052547259
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.6585052013397217,
        "normalized": 0.5295818191011692,
        "contribution": 0.07943727286517538
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.573090076446533,
        "normalized": 0.5481916205507379,
        "contribution": 0.05481916205507379
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8b2c2dd8d107456a43a8164c88525324dd2038c74af1e640f7b922afdc8a8cd0",
    "id": "037e74057049d61a53bb32aa4094",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 5.77639102935791,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.011911727488040924,
    "fitness": 0.8082425722862245,
    "legacy_fitness": 0.8082425722862245,
    "target_cdf": 0.9980776169667565,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.214492321014404,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.830477707644188,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.214492321014404,
        "normalized": 0.9980776169667565,
        "contribution": 0.4990388084833782
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.334988594055176,
        "normalized": 0.3437155282365555,
        "contribution": 0.08592888205913887
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.207131385803223,
        "normalized": 0.989624923740222,
        "contribution": 0.1484437385610333
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.566519260406494,
        "normalized": 0.9706627854063763,
        "contribution": 0.09706627854063764
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/03/037e74057049d61a53bb32aa4094.mp3",
    "audio_sha256": "c5fcde0b61f66129022f112dd88f75e7f35baecd712d6ec48f50cd0aed51e406",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "86fa7d2c814f365812dc7df922ed0c7964afce1379248982a373d79c769b6fba",
    "id": "039c6711093a6c2dc1d30fd863c5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.15364527702331543,
    "fitness": 0.8093327669697051,
    "legacy_fitness": 0.8093327669697051,
    "target_cdf": 0.9371461924705993,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8022687435150146,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7736657878649573,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8022687435150146,
        "normalized": 0.9371461924705993,
        "contribution": 0.46857309623529964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.345165729522705,
        "normalized": 0.3480371802155995,
        "contribution": 0.08700929505389987
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.872772216796875,
        "normalized": 0.8729869940654369,
        "contribution": 0.13094804910981553
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7128210067749023,
        "normalized": 0.8713534746594227,
        "contribution": 0.08713534746594227
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b8b3fc3c32538613f3eba08f400c94a5b853fd7ca3866f0a0ae07a1e432fbeb4",
    "id": "03b2b33722bd245493bd6382b356",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.193545341491699,
    "wer": 0.0,
    "identity": 0.7282589077949524,
    "voiceclap": -0.016285497695207596,
    "fitness": 0.8297648463672057,
    "legacy_fitness": 0.8297648463672057,
    "target_cdf": 0.9974200236848305,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1170973777771,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7297474314617001,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1170973777771,
        "normalized": 0.9974200236848305,
        "contribution": 0.49871001184241526
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4699242115020752,
        "normalized": 0.08099086155600477,
        "contribution": 0.020247715389001192
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.130748271942139,
        "normalized": 0.9142167170262676,
        "contribution": 0.13713250755394013
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1495563983917236,
        "normalized": 0.7365719667634352,
        "contribution": 0.07365719667634352
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f73d944b4c184c4306075c4bdbadf1af9d019ff5be4f1573490c0ffd197da50a",
    "id": "03bfeb1c95688364c592d83d7338",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.499174118041992,
    "genuineness": 4.099404335021973,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06404904276132584,
    "fitness": 0.7866051917696012,
    "legacy_fitness": 0.7866051917696012,
    "target_cdf": 0.9066346497186051,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5831305980682373,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6715406865356328,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5831305980682373,
        "normalized": 0.9066346497186051,
        "contribution": 0.45331732485930254
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.463568687438965,
        "normalized": 0.7826722604742027,
        "contribution": 0.1174008390711304
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8921267986297607,
        "normalized": 0.6570028065606865,
        "contribution": 0.06570028065606866
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "66bb2216b20a62705c6d46c30765b19b1191b5e7f3e904bb28059bc2b6313c89",
    "id": "03fe6c708938afea09ea9833505f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.109982967376709,
    "genuineness": 3.773050308227539,
    "wer": 0.0,
    "identity": 0.2713828682899475,
    "voiceclap": 0.06520767509937286,
    "fitness": 0.8097533682777898,
    "legacy_fitness": 0.8097533682777898,
    "target_cdf": 0.9995867291677825,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.6852335929870605,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8220641543088009,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.6852335929870605,
        "normalized": 0.9995867291677825,
        "contribution": 0.49979336458389123
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2298238277435303,
        "normalized": 0.3003348835264067,
        "contribution": 0.07508372088160167
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.491794586181641,
        "normalized": 0.9948039170931102,
        "contribution": 0.14922058756396653
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.742539405822754,
        "normalized": 0.9796648127934146,
        "contribution": 0.09796648127934146
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "efc622f60a1ce12c7cd1715697496758a94260febc9f23af958e318347780526",
    "id": "046127fe851042273c3e39786e72",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": 0.314607173204422,
    "voiceclap": -0.05321907252073288,
    "fitness": 0.8349280024058763,
    "legacy_fitness": 0.8349280024058763,
    "target_cdf": 0.9867481280341037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5164966583251953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8094403449106768,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5164966583251953,
        "normalized": 0.9867481280341037,
        "contribution": 0.49337406401705186
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3706088066101074,
        "normalized": 0.35892704667845154,
        "contribution": 0.08973176166961289
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.228601455688477,
        "normalized": 0.9269054845040974,
        "contribution": 0.1390358226756146
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7216219902038574,
        "normalized": 0.8729869654839748,
        "contribution": 0.08729869654839749
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4e294f1be1b9e9bcc6a3f16df83b486f7c523a14fef28905f03d56a6ab044e6a",
    "id": "04c5f0576b26a2e09095c7a2ccd0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.229123115539551,
    "genuineness": 4.344170093536377,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": -0.09519477188587189,
    "fitness": 0.776061919837367,
    "legacy_fitness": 0.776061919837367,
    "target_cdf": 0.9321750211493136,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.761687755584717,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7248646259130026,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.761687755584717,
        "normalized": 0.9321750211493136,
        "contribution": 0.4660875105746568
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.046633720397949,
        "normalized": 0.23136007854264456,
        "contribution": 0.05784001963566114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6548268795013428,
        "normalized": 0.828745318479122,
        "contribution": 0.1243117977718683
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2551684379577637,
        "normalized": 0.7662529793081632,
        "contribution": 0.07662529793081632
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bb2c2347d42b15b84b225c24a06e0267d266792b4222b9c7dfc0a92327563562",
    "id": "05001919dcd0d4dfaf6734c18c2d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 4.3881025314331055,
    "wer": 0.0,
    "identity": 0.697882890701294,
    "voiceclap": -0.0005515217781066895,
    "fitness": 0.8360597804843066,
    "legacy_fitness": 0.8360597804843066,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7208259500276628,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.490278720855713,
        "normalized": 0.08456572900066728,
        "contribution": 0.02114143225016682
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9617295265197754,
        "normalized": 0.8885304554003988,
        "contribution": 0.1332795683100598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.030742645263672,
        "normalized": 0.7010225382962612,
        "contribution": 0.07010225382962612
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6d45bfe9dcfaec53805ee3a7286d585d6d5f1ca2e864fc4988d0db6d4bd0a6bf",
    "id": "059534f799678828b1dcd1e977bd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": 0.7054727673530579,
    "voiceclap": 0.03268346190452576,
    "fitness": 0.7575033911612563,
    "legacy_fitness": 0.7575033911612563,
    "target_cdf": 0.8855360207054879,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4613871574401855,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.673409516457118,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4613871574401855,
        "normalized": 0.8855360207054879,
        "contribution": 0.44276801035274393
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5525259971618652,
        "normalized": 0.8049331728558216,
        "contribution": 0.12073997592837322
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0351431369781494,
        "normalized": 0.7023772084492015,
        "contribution": 0.07023772084492015
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "55e8e9584de58cc04451838133d603469f30663b72fdd186e715ae237ac91870",
    "id": "05af0d4dbc26a9e596f21b447a40",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": 0.6912720799446106,
    "voiceclap": -0.14901871979236603,
    "fitness": 0.8309965270483546,
    "legacy_fitness": 0.8309965270483546,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7372223899371213,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8702285289764404,
        "normalized": 0.17425071298327421,
        "contribution": 0.043562678245818554
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7571277618408203,
        "normalized": 0.8506121673397089,
        "contribution": 0.12759182510095632
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2551684379577637,
        "normalized": 0.7662529793081632,
        "contribution": 0.07662529793081632
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0faa544e7f0e9396a5a345f38206ee269a7145796c03af6da2fb7175a52c5d43",
    "id": "05cd29cf84c6a49afb5f2690df96",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.018252063542604446,
    "fitness": 0.8269699046619747,
    "legacy_fitness": 0.8269699046619747,
    "target_cdf": 0.9964005449628859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.003470420837402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.802474718113077,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.003470420837402,
        "normalized": 0.9964005449628859,
        "contribution": 0.49820027248144294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2060770988464355,
        "normalized": 0.29089184034451254,
        "contribution": 0.07272296008612814
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.290871620178223,
        "normalized": 0.9341987553238786,
        "contribution": 0.1401298132985818
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.976851463317871,
        "normalized": 0.9142167224692412,
        "contribution": 0.09142167224692413
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0b6b065d0c9d955e1cfc242b9f20fcc0114922b6e7771d1c9135c5e774ee5b5b",
    "id": "0641c2c65b4b1ac418990d5dd13b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.483288764953613,
    "genuineness": 3.860914945602417,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.06773203611373901,
    "fitness": 0.8212364026865671,
    "legacy_fitness": 0.8212364026865671,
    "target_cdf": 0.9965654549919187,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.019702911376953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7677749252417794,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.019702911376953,
        "normalized": 0.9965654549919187,
        "contribution": 0.49828272749595937
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.768456220626831,
        "normalized": 0.1457841285203917,
        "contribution": 0.03644603213009792
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.67338752746582,
        "normalized": 0.9673901905993718,
        "contribution": 0.14510852858990578
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.756826162338257,
        "normalized": 0.8793763702581625,
        "contribution": 0.08793763702581625
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cbd64a4baefc16a2b35145672838311a62f8cfb195df63bc81d3fa4d0727fafd",
    "id": "0643915905273384e5744514ace5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.482243061065674,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": -0.03392142802476883,
    "fitness": 0.8115370957366207,
    "legacy_fitness": 0.8115370957366207,
    "target_cdf": 0.9970202380692827,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.0684003829956055,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.738138329085756,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.0684003829956055,
        "normalized": 0.9970202380692827,
        "contribution": 0.4985101190346414
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5377724170684814,
        "normalized": 0.09336535255322709,
        "contribution": 0.023341338138306772
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.344245910644531,
        "normalized": 0.9399897500962073,
        "contribution": 0.1409984625144311
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2067627906799316,
        "normalized": 0.7528840939837675,
        "contribution": 0.07528840939837676
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "57725387ac8ee3054c486ead0bd9b188458540a86b7256e33b5a0c4120dd939a",
    "id": "06704a0f38168654228c24f57947",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.061748504638672,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0392584353685379,
    "fitness": 0.8260071671323921,
    "legacy_fitness": 0.8260071671323921,
    "target_cdf": 0.9985049319896988,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.295654773712158,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7622617531498687,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.295654773712158,
        "normalized": 0.9985049319896988,
        "contribution": 0.4992524659948494
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.707392930984497,
        "normalized": 0.130294529860791,
        "contribution": 0.03257363246519775
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.495473384857178,
        "normalized": 0.9542463876094935,
        "contribution": 0.14313695814142402
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7216219902038574,
        "normalized": 0.8729869654839748,
        "contribution": 0.08729869654839749
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ba83103899651ef99d75ce65b6f54a1ff25f452102a7d685528de8de6c074bd9",
    "id": "06955f0c5a31f2765c39a49daa41",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.11995115131139755,
    "fitness": 0.8146638056581293,
    "legacy_fitness": 0.8146638056581293,
    "target_cdf": 0.999062582044815,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.441746711730957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8419873171338471,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.441746711730957,
        "normalized": 0.999062582044815,
        "contribution": 0.4995312910224075
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.433368444442749,
        "normalized": 0.3862667640378337,
        "contribution": 0.09656669100945843
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.385046005249023,
        "normalized": 0.9932207029626189,
        "contribution": 0.14898310544439283
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.540116310119629,
        "normalized": 0.9690622965758837,
        "contribution": 0.09690622965758838
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/06/06955f0c5a31f2765c39a49daa41.mp3",
    "audio_sha256": "07d196dd76b77ed810cba94221a16836782a2f11ceeb1c3c73dc69ff68c28468",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4a67d3cd26fab12fa649ee705e8e46473cbf9db87bfb2679654624e377893888",
    "id": "06eaef17c127b89606cb39e881e6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.5941832065582275,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09041263908147812,
    "fitness": 0.8082484139709412,
    "legacy_fitness": 0.8082484139709412,
    "target_cdf": 0.9979798622347391,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1982598304748535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7920421667020279,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1982598304748535,
        "normalized": 0.9979798622347391,
        "contribution": 0.49898993111736956
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.005924940109253,
        "normalized": 0.21732775018194794,
        "contribution": 0.054331937545486986
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.851302146911621,
        "normalized": 0.9772498567468411,
        "contribution": 0.14658747851202616
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.029657363891602,
        "normalized": 0.921328195271452,
        "contribution": 0.09213281952714521
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/06/06eaef17c127b89606cb39e881e6.mp3",
    "audio_sha256": "57377f8f3ecf38eed9d6c2abe369705d925ddaa6ff8561ed020449da35aca4d7",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "46df8afa8570517a22e53a4c64999abcf06eaf8c22481843f1714cb364e6072c",
    "id": "075ff43ca16083970643d5c69656",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.7793264389038086,
    "wer": 0.0,
    "identity": 0.4155624806880951,
    "voiceclap": -0.08470705151557922,
    "fitness": 0.8083741348843827,
    "legacy_fitness": 0.8083741348843827,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7395340501068589,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.210809707641602,
        "normalized": 0.9247119875768577,
        "contribution": 0.13870679813652864
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.43558931350708,
        "normalized": 0.8123848440876791,
        "contribution": 0.08123848440876791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c9d9ebdd39aa23d5173c4847f0eacc244400e015ceb50f50a6a9909f35c88937",
    "id": "076c764b123a0499ce1330a84ad9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": 0.5907636880874634,
    "voiceclap": 0.008269927464425564,
    "fitness": 0.841363133490662,
    "legacy_fitness": 0.841363133490662,
    "target_cdf": 0.998717748126547,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.344351768493652,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7362539745834709,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.344351768493652,
        "normalized": 0.998717748126547,
        "contribution": 0.4993588740632735
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.476709008216858,
        "normalized": 0.08216957897144506,
        "contribution": 0.020542394742861264
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.424307346343994,
        "normalized": 0.9479187150078625,
        "contribution": 0.14218780725117938
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.167158365249634,
        "normalized": 0.741648985261567,
        "contribution": 0.0741648985261567
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/07/076c764b123a0499ce1330a84ad9.mp3",
    "audio_sha256": "42769e82381e68e01737913ea1e2d84226217e0769139994649b468a0eb820d5",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d852877322942a530459b4560ee5de9d194beca7bd280843b503473d76ed5321",
    "id": "078ead0096a0110faa4e3510a5f6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.425758361816406,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.22643932700157166,
    "fitness": 0.7884625602311822,
    "legacy_fitness": 0.7884625602311822,
    "target_cdf": 0.8855360207054879,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4613871574401855,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6481443865273864,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4613871574401855,
        "normalized": 0.8855360207054879,
        "contribution": 0.44276801035274393
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8363044261932373,
        "normalized": 0.16439289419288416,
        "contribution": 0.04109822354822104
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.316789150238037,
        "normalized": 0.7429102707541737,
        "contribution": 0.11143654061312605
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5169835090637207,
        "normalized": 0.5284161201329535,
        "contribution": 0.05284161201329535
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/07/078ead0096a0110faa4e3510a5f6.mp3",
    "audio_sha256": "d7b4c178acbed435db95f93954701657b26ed3bdae27e7ed8c92ceef5663832e",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7c326076b7cb8a7c9bc54c39bc4cd76b476dab5cea3be0334f1cbef43d8d1e73",
    "id": "07a11933a0dd1a01264e6694afe4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.055472373962402,
    "wer": 0.0,
    "identity": 0.3956599831581116,
    "voiceclap": -0.0125266183167696,
    "fitness": 0.8288076412906809,
    "legacy_fitness": 0.8288076412906809,
    "target_cdf": 0.9978776161334045,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.182027339935303,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7673676704651544,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.182027339935303,
        "normalized": 0.9978776161334045,
        "contribution": 0.49893880806670227
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8396968841552734,
        "normalized": 0.16536210020893177,
        "contribution": 0.04134052505223294
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.353141784667969,
        "normalized": 0.9409148986309699,
        "contribution": 0.1411372347946455
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.651214122772217,
        "normalized": 0.8595110255157367,
        "contribution": 0.08595110255157368
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d4e50b02b1356f169a1f7fc4c36afc6b067a9f7c0fa8e9106280123720271746",
    "id": "07d1ce9e946fd637dbbaa1eb088f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.074300289154053,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06032080948352814,
    "fitness": 0.7783490971551686,
    "legacy_fitness": 0.7783490971551686,
    "target_cdf": 0.8985686055474182,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.534433364868164,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7004951587249133,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.534433364868164,
        "normalized": 0.8985686055474182,
        "contribution": 0.4492843027737091
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9041526317596436,
        "normalized": 0.18447617999078308,
        "contribution": 0.04611904499769577
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7037534713745117,
        "normalized": 0.8394469854437661,
        "contribution": 0.1259170478165649
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3519797325134277,
        "normalized": 0.7917476313694349,
        "contribution": 0.0791747631369435
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2c96bb49a03a4531a948b2a5c96a2424874c4e094b7f5ddc4e5ff622993a0768",
    "id": "07f0997bda98cca1774aa810d67c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": 0.4793047308921814,
    "voiceclap": 0.026628077030181885,
    "fitness": 0.8151665953537346,
    "legacy_fitness": 0.8151665953537346,
    "target_cdf": 0.9673902089874984,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.126917839050293,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7752252035968191,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.126917839050293,
        "normalized": 0.9673902089874984,
        "contribution": 0.4836951044937492
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.302760601043701,
        "normalized": 0.33016657410867467,
        "contribution": 0.08254164352716867
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5525259971618652,
        "normalized": 0.8049331728558216,
        "contribution": 0.12073997592837322
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.774428367614746,
        "normalized": 0.8824847964752807,
        "contribution": 0.08824847964752808
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "91c763dcf13493ec7afd20bfbedc18f6f66aa355e560a659ecac4bcedc0fd2f1",
    "id": "0822916f756591c9325703df16c7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.029954545199871063,
    "fitness": 0.8217673720906722,
    "legacy_fitness": 0.8217673720906722,
    "target_cdf": 0.9983456496378873,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.263189792633057,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8064480022709585,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.263189792633057,
        "normalized": 0.9983456496378873,
        "contribution": 0.49917282481894365
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1450138092041016,
        "normalized": 0.26726902011026354,
        "contribution": 0.06681725502756589
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.833510875701904,
        "normalized": 0.9763929720149331,
        "contribution": 0.14645894580223998
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.188076019287109,
        "normalized": 0.9399897662220895,
        "contribution": 0.09399897662220896
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0bb11f3ac591c6d37eaeaf48961a6731fb5ea431d0e84f7dd3ea6d8e2ed28606",
    "id": "082a334b75184b06c6298b2c9089",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.080576419830322,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.08515419065952301,
    "fitness": 0.7729235924545512,
    "legacy_fitness": 0.7729235924545512,
    "target_cdf": 0.8840175146388185,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.45327091217041,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6924299587559088,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.45327091217041,
        "normalized": 0.8840175146388185,
        "contribution": 0.44200875731940925
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1178743839263916,
        "normalized": 0.25708971381132506,
        "contribution": 0.06427242845283127
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.34792423248291,
        "normalized": 0.7516488247877673,
        "contribution": 0.1127473237181651
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1407554149627686,
        "normalized": 0.7340144926550323,
        "contribution": 0.07340144926550322
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6c88cf22caf8bc8ef385b69e4b9beaa455e8d5cc153505abd0f4fa4f025e9dbd",
    "id": "0871ba182e69913b150493e1e731",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": 0.2334107905626297,
    "voiceclap": 0.010210482403635979,
    "fitness": 0.8202479389940176,
    "legacy_fitness": 0.8202479389940176,
    "target_cdf": 0.9982603445916324,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.246957302093506,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7759805906441726,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.246957302093506,
        "normalized": 0.9982603445916324,
        "contribution": 0.4991301722958162
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.917722225189209,
        "normalized": 0.18866878339787096,
        "contribution": 0.04716719584946774
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.415411949157715,
        "normalized": 0.9470811189066717,
        "contribution": 0.14206216783600076
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7392241954803467,
        "normalized": 0.8762105466288791,
        "contribution": 0.08762105466288791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e1067deaeca26e7ed629deddc4e6dd38217a6b1a177e846ad313c8eb26080aea",
    "id": "08797be11a51c6c21e42f6bf5398",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.642141819000244,
    "genuineness": 4.218649387359619,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10770049691200256,
    "fitness": 0.7944927311662804,
    "legacy_fitness": 0.7944927311662804,
    "target_cdf": 0.905324255769247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.575014352798462,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6579806462055785,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.575014352798462,
        "normalized": 0.905324255769247,
        "contribution": 0.4526621278846235
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6666839122772217,
        "normalized": 0.12062361006755057,
        "contribution": 0.030155902516887642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.34792423248291,
        "normalized": 0.7516488247877673,
        "contribution": 0.1127473237181651
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7931153774261475,
        "normalized": 0.6241529208590221,
        "contribution": 0.06241529208590221
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/08/08797be11a51c6c21e42f6bf5398.mp3",
    "audio_sha256": "c1a42dd69f8903d8b48857ff4ac174a90fbc3c3698c1d9a59dd8cd89c87147c9",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3ce12676953d7cb9f6ec08ae42d641aa96174970117507550f9886e1af2346ac",
    "id": "096176a2b99bad39d19263011953",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11364182084798813,
    "fitness": 0.8326917334074047,
    "legacy_fitness": 0.8326917334074047,
    "target_cdf": 0.9919413812308048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7112860679626465,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7819744329801388,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7112860679626465,
        "normalized": 0.9919413812308048,
        "contribution": 0.4959706906154024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.063595771789551,
        "normalized": 0.23735276923124937,
        "contribution": 0.05933819230781234
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.121852874755859,
        "normalized": 0.9129861397194918,
        "contribution": 0.13694792095792377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.862438201904297,
        "normalized": 0.897176290990003,
        "contribution": 0.0897176290990003
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ff58b3d776f2852452211f076c884cf635f86888a2b71b098d0a1d914063eda7",
    "id": "09d1bb42e02d6318a111e276faf3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.05307499319314957,
    "fitness": 0.8250733068253859,
    "legacy_fitness": 0.8250733068253859,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7742197310058443,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1178743839263916,
        "normalized": 0.25708971381132506,
        "contribution": 0.06427242845283127
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8184332847595215,
        "normalized": 0.890006464053272,
        "contribution": 0.0890006464053272
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "614b6f99c051fbc2a5b090b417c5ac7f45e6226a9d6f91572e39874a22313c56",
    "id": "09dc3962487be4abcf6efbda86ff",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 6.070269584655762,
    "genuineness": 3.7165660858154297,
    "wer": 0.0,
    "identity": 0.5633354187011719,
    "voiceclap": 0.08167359977960587,
    "fitness": 0.8060649184990438,
    "legacy_fitness": 0.8060649184990438,
    "target_cdf": 0.9989025172806805,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.393049240112305,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8775686043382842,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.393049240112305,
        "normalized": 0.9989025172806805,
        "contribution": 0.49945125864034023
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.727447986602783,
        "normalized": 0.5197637231282838,
        "contribution": 0.12994093078207095
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.562960624694824,
        "normalized": 0.9956675509709196,
        "contribution": 0.14935013264563793
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.9889678955078125,
        "normalized": 0.988262822702351,
        "contribution": 0.0988262822702351
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/09/09dc3962487be4abcf6efbda86ff.mp3",
    "audio_sha256": "765e675ec8375af220a70598735d64e4375400fd24896bf4c4f57ccb849c78e0",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "39a34d830df06989bfd648a89a008b628b1c9e6e744ae3837f100ca6e95bdd78",
    "id": "0a361baae3c4263bb15cf82c247c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.055472373962402,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.06859688460826874,
    "fitness": 0.8117768354952852,
    "legacy_fitness": 0.8117768354952852,
    "target_cdf": 0.9673902089874984,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.126917839050293,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7600991298730496,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.126917839050293,
        "normalized": 0.9673902089874984,
        "contribution": 0.4836951044937492
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.158583402633667,
        "normalized": 0.27243382445669295,
        "contribution": 0.06810845611417324
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2639694213867188,
        "normalized": 0.7686398657383282,
        "contribution": 0.07686398657383282
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "714327c96eb465ea1840acdf2ff4d660a0d60e42c75e5e83f7fc3b09a4eab922",
    "id": "0a552e5799cd83770e82d178dacf",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": 0.7862805128097534,
    "voiceclap": -0.023281611502170563,
    "fitness": 0.7208581963718443,
    "legacy_fitness": 0.7208581963718443,
    "target_cdf": 0.7838190157295286,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.0271689891815186,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6179841076422656,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.0271689891815186,
        "normalized": 0.7838190157295286,
        "contribution": 0.3919095078647643
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0602035522460938,
        "normalized": 0.23614752264221534,
        "contribution": 0.059036880660553834
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.192249059677124,
        "normalized": 0.7064243659623586,
        "contribution": 0.10596365489435379
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7535107135772705,
        "normalized": 0.6107406422259369,
        "contribution": 0.0610740642225937
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ef761dcb889101bf2c2cdc847cdd79e3b02c6285c3a24dede00f3615def471a3",
    "id": "0b61556fe3e42c0515a8bdfb4aab",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.05674305558204651,
    "fitness": 0.8242547384815232,
    "legacy_fitness": 0.8242547384815232,
    "target_cdf": 0.9992427094612153,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.50667667388916,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7565444126462021,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.50667667388916,
        "normalized": 0.9992427094612153,
        "contribution": 0.49962135473060765
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6124054193496704,
        "normalized": 0.10853160036418047,
        "contribution": 0.027132900091045117
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.584430694580078,
        "normalized": 0.9612706211599873,
        "contribution": 0.1441905931739981
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6336121559143066,
        "normalized": 0.8559956465055119,
        "contribution": 0.08559956465055119
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "25fa135e80074337a9c1230f20fd6c26787d0148e067219682f6ccc1dd0a59dc",
    "id": "0b7c037adbffd222bb5ac0054a6c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.525146961212158,
    "wer": 0.0,
    "identity": 0.7120879292488098,
    "voiceclap": -0.03916659206151962,
    "fitness": 0.7932198013366027,
    "legacy_fitness": 0.7932198013366027,
    "target_cdf": 0.9772498618564534,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2892422676086426,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7676534961898714,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2892422676086426,
        "normalized": 0.9772498618564534,
        "contribution": 0.4886249309282267
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9652159214019775,
        "normalized": 0.20379983538139917,
        "contribution": 0.050949958845349794
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.362037181854248,
        "normalized": 0.9418287736823874,
        "contribution": 0.1412743160523581
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.695219039916992,
        "normalized": 0.8680429036393686,
        "contribution": 0.08680429036393687
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6e9217dbbd8cb6853a81f94576999b8cd506c964ea8583b8d6ac1ef494b64f37",
    "id": "0ba3b608c4d73116a79fefd9748b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.05058926343917847,
    "fitness": 0.8393829941552018,
    "legacy_fitness": 0.8393829941552018,
    "target_cdf": 0.9887331723771762,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5814266204833984,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7778397521796079,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5814266204833984,
        "normalized": 0.9887331723771762,
        "contribution": 0.4943665861885881
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.046633720397949,
        "normalized": 0.23136007854264456,
        "contribution": 0.05784001963566114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.317558765411377,
        "normalized": 0.9371462024575616,
        "contribution": 0.14057193036863425
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6072089672088623,
        "normalized": 0.8506121598672445,
        "contribution": 0.08506121598672445
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0e82de90b46e5043882f15c588296fff0022244ba52e2b4650c6e4d6b2bee02f",
    "id": "0bc319fcffa123039ce285352481",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 6.1020402908325195,
    "genuineness": 3.9801597595214844,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.11465292423963547,
    "fitness": 0.8176806020070977,
    "legacy_fitness": 0.8176806020070977,
    "target_cdf": 0.9985792095124457,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.311887264251709,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8708478788296007,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.311887264251709,
        "normalized": 0.9985792095124457,
        "contribution": 0.49928960475622286
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.6985063552856445,
        "normalized": 0.506476636234652,
        "contribution": 0.126619159058663
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.029216766357422,
        "normalized": 0.9844679187583991,
        "contribution": 0.14767018781375987
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.601723670959473,
        "normalized": 0.9726892720095494,
        "contribution": 0.09726892720095494
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/0b/0bc319fcffa123039ce285352481.mp3",
    "audio_sha256": "02d8417f53f987fa069a881bbc31beecd3ee1cabb78ba853b81c6ba5212d9183",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "70189e4b5203c49d53035713d5f48ebf662a1e0a245a9548cf2d90a97bcd47d0",
    "id": "0c1332a61be15a9d1d6c4c1e6185",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0760248452425003,
    "fitness": 0.8300479974939725,
    "legacy_fitness": 0.8300479974939725,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7639396486272185,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9686083793640137,
        "normalized": 0.20490763245182997,
        "contribution": 0.05122690811295749
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4619922637939453,
        "normalized": 0.8186299374415724,
        "contribution": 0.08186299374415724
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9c22ce62609d6ff4bb0c62838f04fc95f279ba3880a8146ea37102be1fd87690",
    "id": "0c5531032ac2898bdfb19418b2db",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.250029563903809,
    "wer": 0.0,
    "identity": 0.30437394976615906,
    "voiceclap": -0.09095248579978943,
    "fitness": 0.8271177184981817,
    "legacy_fitness": 0.8271177184981817,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7472956699340906,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9211146831512451,
        "normalized": 0.1897260605487444,
        "contribution": 0.0474315151371861
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.730440616607666,
        "normalized": 0.845096013305189,
        "contribution": 0.12676440199577835
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.475193977355957,
        "normalized": 0.8217031488537243,
        "contribution": 0.08217031488537244
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5a25c5b1de0cba1bf8a6d5011e50853db333696d906a254a6d3350d632901c97",
    "id": "0c809418ceb1eb985d047d82c6fe",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.078212261199951,
    "genuineness": 3.503180742263794,
    "wer": 0.0,
    "identity": 0.6078599691390991,
    "voiceclap": 0.09601083397865295,
    "fitness": 0.796500825908748,
    "legacy_fitness": 0.796500825908748,
    "target_cdf": 0.9971593102355156,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.084632873535156,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8082950613536667,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.084632873535156,
        "normalized": 0.9971593102355156,
        "contribution": 0.4985796551177578
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1212668418884277,
        "normalized": 0.25835102183826425,
        "contribution": 0.06458775545956606
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.278297424316406,
        "normalized": 0.9912255261486401,
        "contribution": 0.148683828922296
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.469708442687988,
        "normalized": 0.9644382185404679,
        "contribution": 0.09644382185404679
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/0c/0c809418ceb1eb985d047d82c6fe.mp3",
    "audio_sha256": "f8ef24b982b7f931ebce268b656c2c918702f0770250d6eb0d9909f10b22819f",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4d7d3429a5d46078ab417408410ed758fc203e008ab5aa0f2d931c8ab6707a54",
    "id": "0cd132542d192f98a7260e5959c8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.948779582977295,
    "wer": 0.0,
    "identity": 0.667567789554596,
    "voiceclap": -0.16109710931777954,
    "fitness": 0.7649129015870769,
    "legacy_fitness": 0.7649129015870769,
    "target_cdf": 0.8793763603106894,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.428922176361084,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6497167278752793,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.428922176361084,
        "normalized": 0.8793763603106894,
        "contribution": 0.4396881801553447
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5035994052886963,
        "normalized": 0.7928660692387137,
        "contribution": 0.11892991038580705
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7381091117858887,
        "normalized": 0.6054879932877197,
        "contribution": 0.060548799328771975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9673cc7c0a61627a51bb41ee3eee929590b210362b0053a20b0992c944c3650e",
    "id": "0d3a51ab972b64b96a1becf2f102",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1457304060459137,
    "fitness": 0.8285266632542349,
    "legacy_fitness": 0.8285266632542349,
    "target_cdf": 0.9929189433068706,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.759983539581299,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7595028136409048,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.759983539581299,
        "normalized": 0.9929189433068706,
        "contribution": 0.4964594716534353
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8227348327636719,
        "normalized": 0.16055304509332224,
        "contribution": 0.04013826127333056
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5984079837799072,
        "normalized": 0.8487882013251657,
        "contribution": 0.08487882013251657
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "eea9790ec9729ef5356f3e0dafa2b77180751b89aaa0c9fa74cb602352048eb0",
    "id": "0d586dceb1a1dbc16cb61e139f1c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.557555198669434,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.14632228016853333,
    "fitness": 0.7924284656531309,
    "legacy_fitness": 0.7924284656531309,
    "target_cdf": 0.8824847618116329,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4451546669006348,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6373091876249831,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4451546669006348,
        "normalized": 0.8824847618116329,
        "contribution": 0.44124238090581647
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2144882678985596,
        "normalized": 0.7131122892698544,
        "contribution": 0.10696684339047816
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.633596897125244,
        "normalized": 0.5693834093138905,
        "contribution": 0.05693834093138905
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "265f0f2691663014e89d9e7ee0fb19bc8161a17f0bc2e2ba3ab8a4b80fbf23be",
    "id": "0db044c679d2d9e83bc40209cc88",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.08641508221626282,
    "fitness": 0.8249669721391066,
    "legacy_fitness": 0.8249669721391066,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7225072670340684,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.826127290725708,
        "normalized": 0.16150748713874752,
        "contribution": 0.04037687178468688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6459312438964844,
        "normalized": 0.826751667074274,
        "contribution": 0.1240127500611411
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.997738838195801,
        "normalized": 0.6907744818671719,
        "contribution": 0.0690774481867172
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/0d/0db044c679d2d9e83bc40209cc88.mp3",
    "audio_sha256": "fa45ddd8be385e349b279b808262044956a7aa0f2fe9eadfdb30dab4b1e05a96",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7f6874dca049037c2a7a4edb8865d2c167e36a37b1886a981900d9435a3f4758",
    "id": "0dfd27cd106c1c4291f4473487bd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.582659721374512,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13887351751327515,
    "fitness": 0.8232852250172449,
    "legacy_fitness": 0.8232852250172449,
    "target_cdf": 0.9409148593413338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.834733486175537,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6577687937689991,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.834733486175537,
        "normalized": 0.9409148593413338,
        "contribution": 0.4704574296706669
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5038483142852783,
        "normalized": 0.08701389328623654,
        "contribution": 0.021753473321559136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5436301231384277,
        "normalized": 0.8027714861707109,
        "contribution": 0.12041572292560664
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.2991585731506348,
        "normalized": 0.45142167851166476,
        "contribution": 0.04514216785116648
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "54632d358a13e09ab5b9524581e21232f5e91c27e576f8f7fea7c3df45d17a69",
    "id": "0e24bb2a8deab7b3c09b9709b33d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": 0.6096416115760803,
    "voiceclap": 0.03652294725179672,
    "fitness": 0.8079449971530897,
    "legacy_fitness": 0.8079449971530897,
    "target_cdf": 0.9585742155738795,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0132906436920166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7347108906447716,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0132906436920166,
        "normalized": 0.9585742155738795,
        "contribution": 0.4792871077869397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9312918186187744,
        "normalized": 0.19291961867711876,
        "contribution": 0.04822990466927969
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8016064167022705,
        "normalized": 0.8595110162463788,
        "contribution": 0.1289266524369568
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3167755603790283,
        "normalized": 0.782672257515953,
        "contribution": 0.07826722575159531
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/0e/0e24bb2a8deab7b3c09b9709b33d.mp3",
    "audio_sha256": "e029620b1b29d8c3ecd0809a4cc668ee1ca5ceb1da122de3cac2937b96e54832",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "31805469ac8781d16baa88a5e6dc57f72b3e723288c6c66be61d2e4529c0ccc3",
    "id": "0ec17fc8641488c384f930f77cf7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.004243344999849796,
    "fitness": 0.8281090015198445,
    "legacy_fitness": 0.8281090015198445,
    "target_cdf": 0.9974200236848305,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1170973777771,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7983872276215936,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1170973777771,
        "normalized": 0.9974200236848305,
        "contribution": 0.49871001184241526
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1110897064208984,
        "normalized": 0.25457682692833283,
        "contribution": 0.06364420673208321
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.548847675323486,
        "normalized": 0.9585742212718315,
        "contribution": 0.1437861331907747
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.038458824157715,
        "normalized": 0.922468758563205,
        "contribution": 0.0922468758563205
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7424b3b6d45122bc82a7925d98f99eaddc8051704f1651fe229681ef8f2406a9",
    "id": "0f088e1e37ad7d67b9cb20d59831",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.036643981933594,
    "wer": 0.0,
    "identity": 0.5985972881317139,
    "voiceclap": 0.001732219010591507,
    "fitness": 0.7794549970134685,
    "legacy_fitness": 0.7794549970134685,
    "target_cdf": 0.8999470925936282,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5425496101379395,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6594242809345081,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5425496101379395,
        "normalized": 0.8999470925936282,
        "contribution": 0.4499735462968141
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.290102005004883,
        "normalized": 0.7352947867084237,
        "contribution": 0.11029421800626356
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8415210247039795,
        "normalized": 0.640342746822993,
        "contribution": 0.0640342746822993
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "de95eac81eb3a960c77ba20a97a63f31d1fdc200e473e0f103e777fe931d0c7c",
    "id": "0f8239ad02600c5ed5ae6060ee0c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.05963090807199478,
    "fitness": 0.8128707750314739,
    "legacy_fitness": 0.8128707750314739,
    "target_cdf": 0.9919413812308048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7112860679626465,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7945845010793479,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7112860679626465,
        "normalized": 0.9919413812308048,
        "contribution": 0.4959706906154024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0873427391052246,
        "normalized": 0.2458838571357379,
        "contribution": 0.061470964283934476
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.726762294769287,
        "normalized": 0.9706627962690428,
        "contribution": 0.14559941944035643
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.985652446746826,
        "normalized": 0.9154342673965461,
        "contribution": 0.09154342673965461
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7bfd85a0e1cd6e991868cd408efc0e8a0adac87bfd86e8437649c7b94a1a6523",
    "id": "0faa485d5b04dca979fb761e8d0b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.049196243286133,
    "wer": 0.0,
    "identity": 0.7063618898391724,
    "voiceclap": -0.13603435456752777,
    "fitness": 0.775738666365617,
    "legacy_fitness": 0.775738666365617,
    "target_cdf": 0.8914684095185468,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.493852138519287,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6679527383228,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.493852138519287,
        "normalized": 0.8914684095185468,
        "contribution": 0.4457342047592734
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4858078956604004,
        "normalized": 0.7883708195307028,
        "contribution": 0.11825562292960541
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8921267986297607,
        "normalized": 0.6570028065606865,
        "contribution": 0.06570028065606866
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fd68b03001731b30970d5e59d05642f7cecd2baaf8954bf0080d9cd5d621d179",
    "id": "0fce300adad97464cd563919defe",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": 0.49632179737091064,
    "voiceclap": 0.07439491152763367,
    "fitness": 0.8223693531613603,
    "legacy_fitness": 0.8223693531613603,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7502530751050088,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.228601455688477,
        "normalized": 0.9269054845040974,
        "contribution": 0.1390358226756146
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4399898052215576,
        "normalized": 0.8134348197593442,
        "contribution": 0.08134348197593443
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "812f6a168fb5f4832ceadcdacf8240b5bb3086b70204540a7b2abdb9b967b5b4",
    "id": "0fe7aaa24f723f0c8c0f3ccae304",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.76,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": 0.6527676582336426,
    "voiceclap": 0.0023385323584079742,
    "fitness": 0.7882866947295988,
    "legacy_fitness": 0.7882866947295988,
    "target_cdf": 0.927983714827443,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7292227745056152,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7233461723682078,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7292227745056152,
        "normalized": 0.927983714827443,
        "contribution": 0.4639918574137215
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9346842765808105,
        "normalized": 0.19399141727969682,
        "contribution": 0.048497854319924205
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3431787490844727,
        "normalized": 0.7895000148839948,
        "contribution": 0.07895000148839948
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5e739cda421d5cb27df3b0425cb7bee6665f2ff2ea76b8d195b8eef998f4140b",
    "id": "1029bf1b31d62de02cebbf034dfd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": 0.7275609970092773,
    "voiceclap": -0.03178655728697777,
    "fitness": 0.7885941490138583,
    "legacy_fitness": 0.7885941490138583,
    "target_cdf": 0.9453737983316208,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.875314712524414,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7361286298833404,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.875314712524414,
        "normalized": 0.9453737983316208,
        "contribution": 0.4726868991658104
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.948253870010376,
        "normalized": 0.19831469683683772,
        "contribution": 0.04957867420920943
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.015103816986084,
        "normalized": 0.8971763038367273,
        "contribution": 0.1345764455755091
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3563802242279053,
        "normalized": 0.7928661093281139,
        "contribution": 0.07928661093281139
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "391f6f6b33ba6d44caff62cfc8fb6cff76d9ddbf23128850e6d9bb46d4665b5c",
    "id": "107fa71edf18fefa0b29a589a00e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.406930446624756,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09047305583953857,
    "fitness": 0.8325655398492071,
    "legacy_fitness": 0.8325655398492071,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7216496609051523,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.810502052307129,
        "normalized": 0.8612466639446046,
        "contribution": 0.1291869995916907
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.960334539413452,
        "normalized": 0.678978307048002,
        "contribution": 0.0678978307048002
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a08c92dc8e89e7e771800a3c940278c24110e0dedd128d78a473dd72f1b07c2e",
    "id": "10c6330e4d87b88778043489126d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.570657730102539,
    "genuineness": 4.381826400756836,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09686584025621414,
    "fitness": 0.7404616377468416,
    "legacy_fitness": 0.7404616377468416,
    "target_cdf": 0.7872381889254867,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.0393433570861816,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6313450707670323,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.0393433570861816,
        "normalized": 0.7872381889254867,
        "contribution": 0.39361909446274335
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3740012645721436,
        "normalized": 0.3603879834715901,
        "contribution": 0.09009699586789753
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.611802577972412,
        "normalized": 0.5132436627030545,
        "contribution": 0.07698654940545818
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.048344612121582,
        "normalized": 0.7064243103093326,
        "contribution": 0.07064243103093326
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "73595291cf31777f1304563f010c047462a024ecc9e26aa863bd92e278cca76e",
    "id": "10dd203132b8a5fafa866b114894",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 4.20609712600708,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": 0.011962473392486572,
    "fitness": 0.7999045785316784,
    "legacy_fitness": 0.7999045785316784,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.761606401854769,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.388724327087402,
        "normalized": 0.9445039636441842,
        "contribution": 0.14167559454662762
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.580805778503418,
        "normalized": 0.8450959649812083,
        "contribution": 0.08450959649812084
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4a9b5cdbd118b90d19d7f27ed9825e8e9a2ef5d23f1f23d4abe90f54f7d54459",
    "id": "10ed5e2143bd29ef952e1874a344",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.948779582977295,
    "wer": 0.0,
    "identity": 0.6376445293426514,
    "voiceclap": -0.04054397717118263,
    "fitness": 0.7746892569584494,
    "legacy_fitness": 0.7746892569584494,
    "target_cdf": 0.9013118501748334,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.550665855407715,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6686731826854362,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.550665855407715,
        "normalized": 0.9013118501748334,
        "contribution": 0.4506559250874167
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.768456220626831,
        "normalized": 0.1457841285203917,
        "contribution": 0.03644603213009792
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4502251148223877,
        "normalized": 0.7792110012337312,
        "contribution": 0.11688165018505968
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.861323118209839,
        "normalized": 0.6468957528286187,
        "contribution": 0.06468957528286187
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a097100cf687754b8eab4cd24aca1bc3acc3ceaf5b3cf3e1a9636525ab9fa57a",
    "id": "11aae7fd4c1774849f0dd9ac48a5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.823258876800537,
    "wer": 0.0,
    "identity": 0.6812807321548462,
    "voiceclap": 0.038710467517375946,
    "fitness": 0.7649855889050452,
    "legacy_fitness": 0.7649855889050452,
    "target_cdf": 0.8943502420521354,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.510084629058838,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7178870828253139,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.510084629058838,
        "normalized": 0.8943502420521354,
        "contribution": 0.4471751210260677
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.161975860595703,
        "normalized": 0.27373275044956896,
        "contribution": 0.06843318761239224
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5970046520233154,
        "normalized": 0.8155238285113872,
        "contribution": 0.12232857427670807
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3827831745147705,
        "normalized": 0.7995019991014589,
        "contribution": 0.0799501999101459
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2f1904ab8647ff8dbf34d2b9e16d7b091cd6b1171c5ff47221cb9fe5476e4318",
    "id": "122a4aaebdaeabb00574518c7b40",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 3,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.143336772918701,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.01473025232553482,
    "fitness": 0.7835033961439719,
    "legacy_fitness": 0.7835033961439719,
    "target_cdf": 0.8999470925936282,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5425496101379395,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7136122377523532,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5425496101379395,
        "normalized": 0.8999470925936282,
        "contribution": 0.4499735462968141
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.10091233253479,
        "normalized": 0.2508316293237907,
        "contribution": 0.06270790733094768
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4858078956604004,
        "normalized": 0.7883708195307028,
        "contribution": 0.11825562292960541
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4971961975097656,
        "normalized": 0.8267516119498604,
        "contribution": 0.08267516119498604
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/12/122a4aaebdaeabb00574518c7b40.mp3",
    "audio_sha256": "572a5abeafff542efdca40bda7ea8c2d364509983d30c4c524d8f31b8da7a86a",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7ab726e340203b0c012da4775dc92a9fca9376327847547ed85afbf3ca399737",
    "id": "12887f6fec7b7c72c961c64e6f67",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.68307638168335,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": -0.08715939521789551,
    "fitness": 0.776218799367813,
    "legacy_fitness": 0.776218799367813,
    "target_cdf": 0.9026628930794968,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.558781862258911,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6634660319783964,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.558781862258911,
        "normalized": 0.9026628930794968,
        "contribution": 0.4513314465397484
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9720008373260498,
        "normalized": 0.206019001985337,
        "contribution": 0.05150475049633425
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.067708730697632,
        "normalized": 0.6677082492667479,
        "contribution": 0.10015623739001218
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7359087467193604,
        "normalized": 0.6047359755230154,
        "contribution": 0.06047359755230154
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "576e30f500c909a46266bc5eef7fee621c047bfbf25392b33ac5b0281f248feb",
    "id": "12c2999dd69823fad59b81927288",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": 0.4771311283111572,
    "voiceclap": 0.07362417876720428,
    "fitness": 0.8154743393530127,
    "legacy_fitness": 0.8154743393530127,
    "target_cdf": 0.9971593102355156,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.084632873535156,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7851062944879121,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.084632873535156,
        "normalized": 0.9971593102355156,
        "contribution": 0.4985796551177578
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9550387859344482,
        "normalized": 0.20049800895093267,
        "contribution": 0.05012450223773317
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.602221965789795,
        "normalized": 0.9625641895690078,
        "contribution": 0.14438462843535116
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.020856857299805,
        "normalized": 0.9201750869706999,
        "contribution": 0.09201750869707
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/12/12c2999dd69823fad59b81927288.mp3",
    "audio_sha256": "d5803246278e2d6954da7e1347e12b4b7233c5942960cd9139e35d87ba3c046f",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4e062cd947d4ec8fa8447f3729578a01ea85d8c304af37c3861405b462ace2ee",
    "id": "13170a17d199fa754b37ec588fea",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.117925643920898,
    "genuineness": 4.2249250411987305,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": -0.05182063207030296,
    "fitness": 0.7705052171492484,
    "legacy_fitness": 0.7705052171492484,
    "target_cdf": 0.9361753561737703,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7941524982452393,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7600337698442249,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7941524982452393,
        "normalized": 0.9361753561737703,
        "contribution": 0.4680876780868852
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3672165870666504,
        "normalized": 0.35746827293550454,
        "contribution": 0.08936706823387613
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6548268795013428,
        "normalized": 0.828745318479122,
        "contribution": 0.1243117977718683
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3167755603790283,
        "normalized": 0.782672257515953,
        "contribution": 0.07826722575159531
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6c18c9384f05f176ab136195de3820bab5a0a7f9d0f2634205e3b29d31102556",
    "id": "1317d0d7ff971fa215666e6d6a0c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.425758361816406,
    "wer": 0.0,
    "identity": 0.6025183200836182,
    "voiceclap": -0.1393873542547226,
    "fitness": 0.8355428421896545,
    "legacy_fitness": 0.8355428421896545,
    "target_cdf": 0.9896249303026692,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6138916015625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.702896351877908,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6138916015625,
        "normalized": 0.9896249303026692,
        "contribution": 0.4948124651513346
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.361367106437683,
        "normalized": 0.06382463962327856,
        "contribution": 0.01595615990581964
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.854980707168579,
        "normalized": 0.8697054717721656,
        "contribution": 0.13045582076582485
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7711129188537598,
        "normalized": 0.6167190605492896,
        "contribution": 0.06167190605492896
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/resolve/main/audio/13/1317d0d7ff971fa215666e6d6a0c.mp3",
    "audio_sha256": "8901a599a5f90794ea7aa5ddd424ec6c00147bba436897a13a18f8df37311d2f",
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6dbfc07e0bd5ee1566a532ea867d6054686ea618036e987a19ff2286deb7ee65",
    "id": "133a4345941113df9dacde36460e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": 0.4942988455295563,
    "voiceclap": -0.1576915681362152,
    "fitness": 0.8298051255667263,
    "legacy_fitness": 0.8298051255667263,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7192231576962881,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.84308922290802,
        "normalized": 0.16633496146433524,
        "contribution": 0.04158374036608381
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.650379180908203,
        "normalized": 0.8277503599224988,
        "contribution": 0.12416255398837481
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8415210247039795,
        "normalized": 0.640342746822993,
        "contribution": 0.0640342746822993
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "72e2ff2fa5382f0a064e43a92338d1a23ff0bd1ee85ef91cd3e3d816e25a0655",
    "id": "135c06bcfd8b19dd4802106db6db",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 3.747946262359619,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.005074646323919296,
    "fitness": 0.8089228803137396,
    "legacy_fitness": 0.8089228803137396,
    "target_cdf": 0.98254389250665,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.402869701385498,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7162103227842288,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.402869701385498,
        "normalized": 0.98254389250665,
        "contribution": 0.491271946253325
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.3952912092208862,
        "normalized": 0.06885472951265481,
        "contribution": 0.017213682378163703
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.121852874755859,
        "normalized": 0.9129861397194918,
        "contribution": 0.13694792095792377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0527453422546387,
        "normalized": 0.7077677319481637,
        "contribution": 0.07077677319481637
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "450911e901fab99073999d15db93b3f7573dc858b4125db142eff53a78b5e629",
    "id": "13ff7556fa97365d6ea97bc5f98a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.181467056274414,
    "genuineness": 4.362998008728027,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.020206106826663017,
    "fitness": 0.8341620351010618,
    "legacy_fitness": 0.8341620351010618,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7789719825317508,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9211146831512451,
        "normalized": 0.1897260605487444,
        "contribution": 0.0474315151371861
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.5310564041137695,
        "normalized": 0.957170068560992,
        "contribution": 0.14357551028414878
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.888841390609741,
        "normalized": 0.9013118211321569,
        "contribution": 0.09013118211321569
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bc93f5583ae7b7953275c1e5500680cb79d39ba2ddbdb9cb7d81492af6521563",
    "id": "141d4d87e7687731a084f1de8fdd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.06122788041830063,
    "fitness": 0.8108174974054649,
    "legacy_fitness": 0.8108174974054649,
    "target_cdf": 0.9557276754241297,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.980825901031494,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7614905778997768,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.980825901031494,
        "normalized": 0.9557276754241297,
        "contribution": 0.47786383771206487
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1450138092041016,
        "normalized": 0.26726902011026354,
        "contribution": 0.06681725502756589
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.952833652496338,
        "normalized": 0.8870402984992682,
        "contribution": 0.13305604477489022
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5456018447875977,
        "normalized": 0.8375344038525582,
        "contribution": 0.08375344038525583
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "579fa4c38e8d3a788c20ac1795d9cac0f3ae2ba46daa77c98cc2ec21852333cf",
    "id": "1476277b16d47c0ea268342ad80d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.998987913131714,
    "wer": 0.0,
    "identity": 0.5702840089797974,
    "voiceclap": -0.009590040892362595,
    "fitness": 0.7786481817279147,
    "legacy_fitness": 0.7786481817279147,
    "target_cdf": 0.9026628930794968,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.558781862258911,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7110807674639068,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.558781862258911,
        "normalized": 0.9026628930794968,
        "contribution": 0.4513314465397484
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0839502811431885,
        "normalized": 0.24465512789027288,
        "contribution": 0.06116378197256822
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5347344875335693,
        "normalized": 0.8005954285570323,
        "contribution": 0.12008931428355483
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3255765438079834,
        "normalized": 0.7849622466803536,
        "contribution": 0.07849622466803537
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "aad934ad46c6ed23d66eb353f5be5d1cf911cee3b6e293560759debba4f71aed",
    "id": "14ed7d5bc7804b1ac3d6a3c1db59",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": 0.533572793006897,
    "voiceclap": 0.03908152133226395,
    "fitness": 0.8203211034735161,
    "legacy_fitness": 0.8203211034735161,
    "target_cdf": 0.999200941205048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.490444183349609,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8214191253805088,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.490444183349609,
        "normalized": 0.999200941205048,
        "contribution": 0.499600470602524
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2400012016296387,
        "normalized": 0.3044239560639862,
        "contribution": 0.07610598901599655
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.207131385803223,
        "normalized": 0.989624923740222,
        "contribution": 0.1484437385610333
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.601723670959473,
        "normalized": 0.9726892720095494,
        "contribution": 0.09726892720095494
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "20e63690a1185285db278d89bb46e23559d051297b7c717be884a8383b1e2b60",
    "id": "1511cc46c3f477f4b7ff0349a655",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.197352409362793,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": 0.5333949327468872,
    "voiceclap": -0.1817331612110138,
    "fitness": 0.807306666799892,
    "legacy_fitness": 0.807306666799892,
    "target_cdf": 0.9427315386702796,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.850965976715088,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7115873991072451,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.850965976715088,
        "normalized": 0.9427315386702796,
        "contribution": 0.4713657693351398
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0296716690063477,
        "normalized": 0.2254527264287831,
        "contribution": 0.05636318160719578
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.365715742111206,
        "normalized": 0.7565702742161262,
        "contribution": 0.11348554113241893
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.039543628692627,
        "normalized": 0.7037290703249057,
        "contribution": 0.07037290703249058
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "32e49a49afd7ef1d50dab406e71d429f1e587a9acf9372f15f5944e3985c2bd6",
    "id": "15239b8ef5233e53bb14ffc89411",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 5.990842819213867,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0397266149520874,
    "fitness": 0.8207552416793045,
    "legacy_fitness": 0.8207552416793045,
    "target_cdf": 0.9987822818103504,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.360584259033203,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7830369530076706,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.360584259033203,
        "normalized": 0.9987822818103504,
        "contribution": 0.4993911409051752
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.917722225189209,
        "normalized": 0.18866878339787096,
        "contribution": 0.04716719584946774
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.6555962562561035,
        "normalized": 0.9662345967558206,
        "contribution": 0.14493518951337309
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.985652446746826,
        "normalized": 0.9154342673965461,
        "contribution": 0.09154342673965461
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fdfe9666c3a48e25e723853edefbbe194931e0cdf05c59cce4dfe95d0726e5f8",
    "id": "15644b0b8f2df244d1238f75c8e5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.961331605911255,
    "wer": 0.0,
    "identity": 0.6317917704582214,
    "voiceclap": -0.13006095588207245,
    "fitness": 0.7638473791547497,
    "legacy_fitness": 0.7638473791547497,
    "target_cdf": 0.8793763603106894,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.428922176361084,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6676987699645586,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.428922176361084,
        "normalized": 0.8793763603106894,
        "contribution": 0.4396881801553447
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8193423748016357,
        "normalized": 0.1596022950986577,
        "contribution": 0.039900573774664425
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.427985668182373,
        "normalized": 0.7733726160960801,
        "contribution": 0.116005892414412
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.096750259399414,
        "normalized": 0.721041236201374,
        "contribution": 0.07210412362013739
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "40b87cea501856795d543f462811d38ab13bca65b3c1fbbc5b4fedbc3f1cca6b",
    "id": "167856aed99765731fb5e1db8701",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": 0.731842041015625,
    "voiceclap": -0.04582254961133003,
    "fitness": 0.7585148659840208,
    "legacy_fitness": 0.7585148659840208,
    "target_cdf": 0.8680429235090776,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3721086978912354,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6761791728093596,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3721086978912354,
        "normalized": 0.8680429235090776,
        "contribution": 0.4340214617545388
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.009317398071289,
        "normalized": 0.21847802587457404,
        "contribution": 0.05461950646864351
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4235379695892334,
        "normalized": 0.7721946200719969,
        "contribution": 0.11582919301079952
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0835487842559814,
        "normalized": 0.7170901157537775,
        "contribution": 0.07170901157537775
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e9a94377870d75694979e102fd866a223bb50c38f7a9be621a805b28609432e5",
    "id": "174c321261c37e582e20da998273",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.44,
    "cfg": 2,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.268857479095459,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.012507492676377296,
    "fitness": 0.8002373579398139,
    "legacy_fitness": 0.8002373579398139,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6407641324924473,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5784813165664673,
        "normalized": 0.10143141696818753,
        "contribution": 0.02535785424204688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2945499420166016,
        "normalized": 0.736571978107565,
        "contribution": 0.11048579671613475
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.2903575897216797,
        "normalized": 0.44832959339945416,
        "contribution": 0.04483295933994542
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6a97230c90673b6fc7a197ad1265956c2fdc3297e78f142db9c95cc0810fd443",
    "id": "18213b9481301e2a347003956f5b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 4.162164688110352,
    "wer": 0.0,
    "identity": 0.4222780764102936,
    "voiceclap": -0.05381465330719948,
    "fitness": 0.8266457036791919,
    "legacy_fitness": 0.8266457036791919,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7794606770691913,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9652159214019775,
        "normalized": 0.20379983538139917,
        "contribution": 0.050949958845349794
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.388724327087402,
        "normalized": 0.9445039636441842,
        "contribution": 0.14167559454662762
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9152445793151855,
        "normalized": 0.9053242803940382,
        "contribution": 0.09053242803940383
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c95645891a3d5e1a800f39f5a16b1ccdaae60d3d797a0f8f73be06d00c47c441",
    "id": "18a26fcdd96fba851088d660c226",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": 0.5268599987030029,
    "voiceclap": 0.12209966778755188,
    "fitness": 0.8204401629197927,
    "legacy_fitness": 0.8204401629197927,
    "target_cdf": 0.988262823627061,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5651941299438477,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.857319235179687,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5651941299438477,
        "normalized": 0.988262823627061,
        "contribution": 0.4941314118135305
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.7092137336730957,
        "normalized": 0.5113940109330486,
        "contribution": 0.12784850273326215
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.442098617553711,
        "normalized": 0.9495623364922869,
        "contribution": 0.14243435047384304
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.091264724731445,
        "normalized": 0.9290497015905134,
        "contribution": 0.09290497015905134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "71bd1256e9d2dde5137a4059d03b495b4ad39976ebc0bdffc4d9620b77b2fc41",
    "id": "197b363f5a0211a5279b2750a07f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.8295347690582275,
    "wer": 0.0,
    "identity": 0.7636817693710327,
    "voiceclap": -0.03133288025856018,
    "fitness": 0.7603279151842842,
    "legacy_fitness": 0.7603279151842842,
    "target_cdf": 0.880937722986112,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4370384216308594,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6400316182330232,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4370384216308594,
        "normalized": 0.880937722986112,
        "contribution": 0.440468861493056
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6870384216308594,
        "normalized": 0.12539399986239474,
        "contribution": 0.031348499965598686
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.223383903503418,
        "normalized": 0.7157670812548709,
        "contribution": 0.10736506218823062
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7469100952148438,
        "normalized": 0.6084919458613794,
        "contribution": 0.060849194586137946
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "435ff192a7e25ee786872ea60fdc484cb6c34ae47ab2c07b17c019b7fffd0828",
    "id": "19e3f70ef64080ef31680ce882d3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.7793264389038086,
    "wer": 0.0,
    "identity": 0.7674958109855652,
    "voiceclap": 0.029248792678117752,
    "fitness": 0.7923221121248603,
    "legacy_fitness": 0.7923221121248603,
    "target_cdf": 0.947918719056646,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8996634483337402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6972391645846919,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8996634483337402,
        "normalized": 0.947918719056646,
        "contribution": 0.473959359528323
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5649117231369019,
        "normalized": 0.0986881865760929,
        "contribution": 0.024672046644023227
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.810502052307129,
        "normalized": 0.8612466639446046,
        "contribution": 0.1291869995916907
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.008740186691284,
        "normalized": 0.6942075882065499,
        "contribution": 0.069420758820655
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "48f5211852a4184a438f5674dfb6b151f8021c80725acad8b8bc7494aa504eb3",
    "id": "1a369587b2cf4a1dbc79d1d4f84e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.532451152801514,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1830286830663681,
    "fitness": 0.8184207298299397,
    "legacy_fitness": 0.8184207298299397,
    "target_cdf": 0.9445039705317666,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8671984672546387,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6965035144275379,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8671984672546387,
        "normalized": 0.9445039705317666,
        "contribution": 0.4722519852658833
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8091652393341064,
        "normalized": 0.1567722644593298,
        "contribution": 0.03919306611483245
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5169429779052734,
        "normalized": 0.7962001195838251,
        "contribution": 0.11943001793757375
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8899264335632324,
        "normalized": 0.6562844510924838,
        "contribution": 0.06562844510924838
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "febfb8765f8fb437cb93d2f194fab51d41bcf43220421d891e7570101128e160",
    "id": "1a42d209f735234005954be1c34f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.149612903594971,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13313238322734833,
    "fitness": 0.8126281780140445,
    "legacy_fitness": 0.8126281780140445,
    "target_cdf": 0.9592621807529222,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.021406888961792,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6924065452473308,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.021406888961792,
        "normalized": 0.9592621807529222,
        "contribution": 0.4796310903764611
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7215447425842285,
        "normalized": 0.8432277384318043,
        "contribution": 0.12648416076477065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8283193111419678,
        "normalized": 0.6359507907076711,
        "contribution": 0.06359507907076711
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9fcb56f6665bdb0e7402a3ba4d784be58764d7c1e815512e8124b8ea674801e4",
    "id": "1a532bfbe2db27367b5d5774bb93",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.1998209953308105,
    "wer": 0.0,
    "identity": 0.7040686011314392,
    "voiceclap": -0.048687055706977844,
    "fitness": 0.8292088347390087,
    "legacy_fitness": 0.8292088347390087,
    "target_cdf": 0.9977706984112247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.165794849395752,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7488119713691302,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.165794849395752,
        "normalized": 0.9977706984112247,
        "contribution": 0.49888534920561234
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6191902160644531,
        "normalized": 0.1099935309435226,
        "contribution": 0.02749838273588065
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.406516075134277,
        "normalized": 0.9462328279462877,
        "contribution": 0.14193492419194315
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.404785633087158,
        "normalized": 0.8049331523569412,
        "contribution": 0.08049331523569413
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6f2818c7dbd4d66a22a79a2e25c36ea528525765f3eabe16dedf602d3f8090d8",
    "id": "1a7ef75f0a62e9d0f2700f93b72b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 5.998785495758057,
    "genuineness": 4.532451152801514,
    "wer": 0.0,
    "identity": 0.722832441329956,
    "voiceclap": -0.1730869561433792,
    "fitness": 0.8033466232452302,
    "legacy_fitness": 0.8033466232452302,
    "target_cdf": 0.929049708969098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7373390197753906,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6774709165513123,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7373390197753906,
        "normalized": 0.929049708969098,
        "contribution": 0.464524854484549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9448615312576294,
        "normalized": 0.19722848067005766,
        "contribution": 0.049307120167514415
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.112187385559082,
        "normalized": 0.6817707678889575,
        "contribution": 0.10226561518334362
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7623119354248047,
        "normalized": 0.6137332671590532,
        "contribution": 0.06137332671590532
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "233776a57513dd3a9587ce4ad42a2718ced8c4828a91e1df584ce69b759b4980",
    "id": "1ad5805d7407dc7ed8ec2042440a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 3.9676074981689453,
    "wer": 0.0,
    "identity": 0.764564037322998,
    "voiceclap": 0.03470531851053238,
    "fitness": 0.8179949781244074,
    "legacy_fitness": 0.8179949781244074,
    "target_cdf": 0.999062582044815,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.441746711730957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.755699386404513,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.441746711730957,
        "normalized": 0.999062582044815,
        "contribution": 0.4995312910224075
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.476709008216858,
        "normalized": 0.08216957897144506,
        "contribution": 0.020542394742861264
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.780136585235596,
        "normalized": 0.9736578749090112,
        "contribution": 0.14604868123635167
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.853637218475342,
        "normalized": 0.8957701940289253,
        "contribution": 0.08957701940289253
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "33a4cdca844b762ccc5df630f4b77adfc72b474443203c272b62b0484d477321",
    "id": "1ad8b634ec3952689322a2756fed",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 3.7667744159698486,
    "wer": 0.0,
    "identity": 0.7058332562446594,
    "voiceclap": 0.011040469631552696,
    "fitness": 0.7926909487320071,
    "legacy_fitness": 0.7926909487320071,
    "target_cdf": 0.956453669244664,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9889419078826904,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7400070649294764,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9889419078826904,
        "normalized": 0.956453669244664,
        "contribution": 0.478226834622332
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8600512742996216,
        "normalized": 0.17125467685227086,
        "contribution": 0.042813669213067715
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.059582710266113,
        "normalized": 0.9040003875053659,
        "contribution": 0.1356000581258049
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5279998779296875,
        "normalized": 0.8336650296827179,
        "contribution": 0.0833665029682718
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f1c24dd31c77b4afc438fb40f791adc68b87c69089553d1c45ba158a14ca30db",
    "id": "1af1a6ba3878bde84dcce6297b6e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.337893962860107,
    "wer": 0.0,
    "identity": 0.6289641857147217,
    "voiceclap": -0.0498926043510437,
    "fitness": 0.8332746538724616,
    "legacy_fitness": 0.8332746538724616,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.684105967814676,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4724645614624023,
        "normalized": 0.7849622943953769,
        "contribution": 0.11774434415930653
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4771039485931396,
        "normalized": 0.5143144151388265,
        "contribution": 0.05143144151388265
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "81f210258d12ee6242e5ef16155bbc154b0346b80c367fd544d3dd7d4bd3d134",
    "id": "1b45b8105d85eb75123c8e95be80",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.660081624984741,
    "wer": 0.0,
    "identity": 0.47626182436943054,
    "voiceclap": 0.0953933522105217,
    "fitness": 0.8101491537539716,
    "legacy_fitness": 0.8101491537539716,
    "target_cdf": 0.999864011376809,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.993649959564209,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8745765748170734,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.993649959564209,
        "normalized": 0.999864011376809,
        "contribution": 0.4999320056884045
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.6899194717407227,
        "normalized": 0.5025323540877488,
        "contribution": 0.1256330885219372
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 6.025538444519043,
        "normalized": 0.9987822819861399,
        "contribution": 0.14981734229792099
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 5.147386074066162,
        "normalized": 0.9919413830881068,
        "contribution": 0.09919413830881069
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9b53df8501eff0ee7b32ddc18860f36e3dd6c15656a6972e03dda64a8fe9a538",
    "id": "1b4dbec4fa8af58d7bedb02d4422",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 4.036643981933594,
    "wer": 0.0,
    "identity": 0.6399446725845337,
    "voiceclap": 0.0742596834897995,
    "fitness": 0.8134833010388387,
    "legacy_fitness": 0.8134833010388387,
    "target_cdf": 0.9711807712620442,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1837315559387207,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7133460976124363,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1837315559387207,
        "normalized": 0.9711807712620442,
        "contribution": 0.4855903856310221
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5920509099960327,
        "normalized": 0.10422977602228944,
        "contribution": 0.02605744400557236
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.917250871658325,
        "normalized": 0.8809377370125135,
        "contribution": 0.13214066055187704
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0131406784057617,
        "normalized": 0.6955760742396476,
        "contribution": 0.06955760742396476
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "feee2bed464dded64ba38170c102b4151e02dbccf8a0ffde85dc52568a00f02b",
    "id": "1b6cb19217c39d76e081c90d0eda",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.007244117558002472,
    "fitness": 0.7819182779358247,
    "legacy_fitness": 0.7819182779358247,
    "target_cdf": 0.9390533101491965,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8185012340545654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7958186283273655,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8185012340545654,
        "normalized": 0.9390533101491965,
        "contribution": 0.46952665507459823
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.387570858001709,
        "normalized": 0.36625184367446406,
        "contribution": 0.09156296091861602
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.3709330558776855,
        "normalized": 0.9427315543087618,
        "contribution": 0.14140973314631425
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b4185d8c2f20f92d1ff57b682c0cf4021b1bae1e11094ff1946712edfe889c10",
    "id": "1b711c36aceba37124b2b2fb3c59",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.515059471130371,
    "genuineness": 4.080576419830322,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.08028179407119751,
    "fitness": 0.8283240701425296,
    "legacy_fitness": 0.8283240701425296,
    "target_cdf": 0.9908471317426806,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6625890731811523,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7303167344844327,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6625890731811523,
        "normalized": 0.9908471317426806,
        "contribution": 0.4954235658713403
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4631394147872925,
        "normalized": 0.07982495177638921,
        "contribution": 0.019956237944097302
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.228601455688477,
        "normalized": 0.9269054845040974,
        "contribution": 0.1390358226756146
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2287654876708984,
        "normalized": 0.7590110799338052,
        "contribution": 0.07590110799338053
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "25908ba004963ad024b0f4b73f0ed8b5683eef1bfc53d4a635d2d10a7c82d4aa",
    "id": "1c636899c8f237e78d7a2ac7d702",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 3.9801597595214844,
    "wer": 0.0,
    "identity": 0.3624517619609833,
    "voiceclap": 0.05459433048963547,
    "fitness": 0.8213055179690267,
    "legacy_fitness": 0.8213055179690267,
    "target_cdf": 0.9998720340281618,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.00988245010376,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7857946053382817,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.00988245010376,
        "normalized": 0.9998720340281618,
        "contribution": 0.4999360170140809
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8363044261932373,
        "normalized": 0.16439289419288416,
        "contribution": 0.04109822354822104
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.2427144050598145,
        "normalized": 0.9904545174589181,
        "contribution": 0.1485681776188377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.43450403213501,
        "normalized": 0.9619218715714208,
        "contribution": 0.09619218715714209
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0c0f9d88cd0a330eeccee15a6dd8d60cf5821b6fc1fcc7c4c85f11cd5df2d1ea",
    "id": "1c802d8089135baf20fd15d6a064",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.133810997009277,
    "genuineness": 3.9425034523010254,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": 0.08622317761182785,
    "fitness": 0.7882164160923207,
    "legacy_fitness": 0.7882164160923207,
    "target_cdf": 0.9995867291677825,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.6852335929870605,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.787772851214672,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.6852335929870605,
        "normalized": 0.9995867291677825,
        "contribution": 0.49979336458389123
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9007601737976074,
        "normalized": 0.18343713514349452,
        "contribution": 0.04585928378587363
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.064799785614014,
        "normalized": 0.9856469755282963,
        "contribution": 0.14784704632924445
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.214478969573975,
        "normalized": 0.9427315651566268,
        "contribution": 0.09427315651566269
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "29b243d20518385d9ea9a126140d1ac769a7f79fa2ebab8b38d991cd33fe5abf",
    "id": "1c9e5231e5eec36aec32043a1f32",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.141753673553467,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.011945823207497597,
    "fitness": 0.8252535002674434,
    "legacy_fitness": 0.8252535002674434,
    "target_cdf": 0.996049248624995,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.971005439758301,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7447532194360474,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.971005439758301,
        "normalized": 0.996049248624995,
        "contribution": 0.4980246243124975
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5456018447875977,
        "normalized": 0.8375344038525582,
        "contribution": 0.08375344038525583
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3a1660fcbf9d3ee58ecb788e2248ee22095b46fea912b9946975658fdf296a41",
    "id": "1d3805737aaa7c85a13511288cef",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": 0.5203723907470703,
    "voiceclap": -0.028975529596209526,
    "fitness": 0.8362268809153901,
    "legacy_fitness": 0.8362268809153901,
    "target_cdf": 0.9919413812308048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7112860679626465,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7465662774681961,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7112860679626465,
        "normalized": 0.9919413812308048,
        "contribution": 0.4959706906154024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9261465072631836,
        "normalized": 0.8824847562300842,
        "contribution": 0.13237271343451262
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2199645042419434,
        "normalized": 0.7565702736046067,
        "contribution": 0.07565702736046068
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4312871129b3e043eb3f9246ca163932c3fac54f5a2098314c161163609bf2e2",
    "id": "1d3bff6813e8062a555934b340f1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.22903618216514587,
    "fitness": 0.8056284917030243,
    "legacy_fitness": 0.8056284917030243,
    "target_cdf": 0.929049708969098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7373390197753906,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6800527692803986,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7373390197753906,
        "normalized": 0.929049708969098,
        "contribution": 0.464524854484549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.383507251739502,
        "normalized": 0.7614385460618895,
        "contribution": 0.11421578190928343
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.933931589126587,
        "normalized": 0.6705405529501507,
        "contribution": 0.06705405529501507
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b44a45fdb42b16dafa9b07c25dada10d6fdc79691293e7054ce78d2b48baaefb",
    "id": "1d9f3422c5145d95f124e319a4c9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.886019229888916,
    "wer": 0.0,
    "identity": 0.5562150478363037,
    "voiceclap": 0.052312351763248444,
    "fitness": 0.755278228581176,
    "legacy_fitness": 0.755278228581176,
    "target_cdf": 0.8697054919803655,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3802249431610107,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7100133234997323,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3802249431610107,
        "normalized": 0.8697054919803655,
        "contribution": 0.43485274599018275
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.216254234313965,
        "normalized": 0.2949218491618698,
        "contribution": 0.07373046229046745
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5792131423950195,
        "normalized": 0.8113311992293513,
        "contribution": 0.12169967988440268
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3739821910858154,
        "normalized": 0.7973043533467947,
        "contribution": 0.07973043533467948
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a8751f0ce093ea66114eac6f49806fb8610749458b941c89e3ac6c6b795eeb92",
    "id": "1dc9019963b5477a90de5fbe8eb1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.109982967376709,
    "genuineness": 4.093128204345703,
    "wer": 0.06666666666666667,
    "identity": 0.36298668384552,
    "voiceclap": -0.05758904293179512,
    "fitness": 0.7592727742331211,
    "legacy_fitness": 0.7592727742331211,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8055728275628453,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2705328464508057,
        "normalized": 0.31683685573425413,
        "contribution": 0.07920921393356353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.415411949157715,
        "normalized": 0.9470811189066717,
        "contribution": 0.14206216783600076
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.196876525878906,
        "normalized": 0.9409148467090185,
        "contribution": 0.09409148467090185
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "545fef26508242eb86b785c4134c671e96df52959a648ba953cf3451ec7e74a0",
    "id": "1df9bf32f4af15811bcef6fddfc2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 3,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 3.886019229888916,
    "wer": 0.0,
    "identity": 0.5718556642532349,
    "voiceclap": -0.0544457882642746,
    "fitness": 0.7953559958623713,
    "legacy_fitness": 0.7953559958623713,
    "target_cdf": 0.9470810897522896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.891547203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7858074756203418,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.891547203063965,
        "normalized": 0.9470810897522896,
        "contribution": 0.4735405448761448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2942795753479004,
        "normalized": 0.3266368847639896,
        "contribution": 0.0816592211909974
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.210809707641602,
        "normalized": 0.9247119875768577,
        "contribution": 0.13870679813652864
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.012055397033691,
        "normalized": 0.91900911416671,
        "contribution": 0.091900911416671
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "20b09ec556ea335b0af286f16fe3f372ead9d32df2a7f8f22123a3fcffc9ab73",
    "id": "1e10ef4bd205231a8ff9bcad49ba",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.8671910762786865,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.05910268798470497,
    "fitness": 0.8191693522041013,
    "legacy_fitness": 0.8191693522041013,
    "target_cdf": 0.9994538858542781,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.604071140289307,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7990101999071977,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.604071140289307,
        "normalized": 0.9994538858542781,
        "contribution": 0.49972694292713904
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.033064126968384,
        "normalized": 0.2266273368759334,
        "contribution": 0.05665683421898335
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.029216766357422,
        "normalized": 0.9844679187583991,
        "contribution": 0.14767018781375987
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.284886837005615,
        "normalized": 0.9495623494731538,
        "contribution": 0.09495623494731538
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3f8771c50c28c3ef3c8bfcb52958c8c8000cd4e57c5b6070e179d6060dc56d91",
    "id": "1e51b697d530819696bef0852944",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.432034492492676,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12530475854873657,
    "fitness": 0.77395384902079,
    "legacy_fitness": 0.77395384902079,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6338712593413172,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2456233501434326,
        "normalized": 0.7223522963184019,
        "contribution": 0.10835284444776029
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7117059230804443,
        "normalized": 0.5964390216648275,
        "contribution": 0.05964390216648276
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7b6ac26ebd9058cb2ae825db261aff3bbd02d880deda4d2ff817246c31bd4dba",
    "id": "1e7e5de56b1b50a744ff573418a9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.157639026641846,
    "genuineness": 4.2249250411987305,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1325288712978363,
    "fitness": 0.8278132940465268,
    "legacy_fitness": 0.8278132940465268,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7305189182430681,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4224305152893066,
        "normalized": 0.07309452831431107,
        "contribution": 0.018273632078577767
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2463674545288086,
        "normalized": 0.7638525058566775,
        "contribution": 0.07638525058566775
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "078d056d298f039bce03f7addba36ed3107bd6ad2a6461e6fcad10ac0eb40bf8",
    "id": "1f035bd426cedb61ad481e633f11",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": 0.723368227481842,
    "voiceclap": 0.09972406923770905,
    "fitness": 0.8258339296446506,
    "legacy_fitness": 0.8258339296446506,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7437023120684809,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.112957000732422,
        "normalized": 0.9117423574294996,
        "contribution": 0.13676135361442493
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3519797325134277,
        "normalized": 0.7917476313694349,
        "contribution": 0.0791747631369435
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c3ce0de95e108faf3d7e017eeeee367f708b7d6d2cb828758614bf09287d5932",
    "id": "1f1fc059eb76d14fbef6e5aa4d7e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.532451152801514,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.22890958189964294,
    "fitness": 0.6731758205892051,
    "legacy_fitness": 0.6731758205892051,
    "target_cdf": 0.7553448661739071,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.929774284362793,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6147712152521727,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.929774284362793,
        "normalized": 0.7553448661739071,
        "contribution": 0.37767243308695353
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2518744468688965,
        "normalized": 0.3092254511509299,
        "contribution": 0.07730636278773248
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.869778871536255,
        "normalized": 0.602477737045545,
        "contribution": 0.09037166055683174
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.008740186691284,
        "normalized": 0.6942075882065499,
        "contribution": 0.069420758820655
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b038bc5a174117ef71cfc0f023dc9915329dcbea51dd035845cedeacdd3a3111",
    "id": "1f2042befb65712b02a2aca0f249",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.3127899169921875,
    "wer": 0.0,
    "identity": 0.6900442838668823,
    "voiceclap": -0.03621483966708183,
    "fitness": 0.748440221726008,
    "legacy_fitness": 0.748440221726008,
    "target_cdf": 0.8113312001607969,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.128621816635132,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5981545987414691,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.128621816635132,
        "normalized": 0.8113312001607969,
        "contribution": 0.4056656000803984
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.174457550048828,
        "normalized": 0.7010225607218077,
        "contribution": 0.10515338410827114
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5114829540252686,
        "normalized": 0.5264728003277405,
        "contribution": 0.05264728003277405
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1a42e3f5599d87087c134ca81a8f0c965ff5c3894f48a5cc0af7dba78e26bae9",
    "id": "1f4fcb13821edb2ad07ceb0ba58d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.973883628845215,
    "wer": 0.0,
    "identity": 0.6783046126365662,
    "voiceclap": -0.01416931301355362,
    "fitness": 0.8256577882874745,
    "legacy_fitness": 0.8256577882874745,
    "target_cdf": 0.9979798622347391,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1982598304748535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7847731551976098,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1982598304748535,
        "normalized": 0.9979798622347391,
        "contribution": 0.49898993111736956
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0602035522460938,
        "normalized": 0.23614752264221534,
        "contribution": 0.059036880660553834
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.677617073059082,
        "normalized": 0.8646740371672423,
        "contribution": 0.08646740371672423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c69407c274cbb8b8b606f791e5a81c71bb85822ce43260b8ffba4bcea772c790",
    "id": "1f6ce197bdf2fd1a3c41961229a7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09992954134941101,
    "fitness": 0.8191805569132515,
    "legacy_fitness": 0.8191805569132515,
    "target_cdf": 0.9952532227437393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9060754776000977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7867104763838875,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9060754776000977,
        "normalized": 0.9952532227437393,
        "contribution": 0.49762661137186964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9889628887176514,
        "normalized": 0.21162916044311025,
        "contribution": 0.05290729011077756
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.833510875701904,
        "normalized": 0.9763929720149331,
        "contribution": 0.14645894580223998
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.862438201904297,
        "normalized": 0.897176290990003,
        "contribution": 0.0897176290990003
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "30d810d7c8cd6b5593f24193306fa35a2b82eba7a1967091a7c62ea993ff4540",
    "id": "1faeb7608349d09b014f3d244083",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1070994883775711,
    "fitness": 0.8316220249236389,
    "legacy_fitness": 0.8316220249236389,
    "target_cdf": 0.9772498618564534,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2892422676086426,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7487355256573446,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2892422676086426,
        "normalized": 0.9772498618564534,
        "contribution": 0.4886249309282267
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.059582710266113,
        "normalized": 0.9040003875053659,
        "contribution": 0.1356000581258049
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.444390296936035,
        "normalized": 0.8144811489247583,
        "contribution": 0.08144811489247583
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "10a942a82e65d9bc7e4f8b96acce8f077e5d9f0a787ccf494cb4306d6fe31324",
    "id": "1fbf3340d19be4f1deb87b908b17",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.6203155517578125,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.054096221923828125,
    "fitness": 0.8477209921338834,
    "legacy_fitness": 0.8477209921338834,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7116723804328984,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4224305152893066,
        "normalized": 0.07309452831431107,
        "contribution": 0.018273632078577767
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.970625400543213,
        "normalized": 0.8900064909716228,
        "contribution": 0.1335009736457434
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8283193111419678,
        "normalized": 0.6359507907076711,
        "contribution": 0.06359507907076711
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ebfb9c2a900884b49b1b155204ca641a87af4a4e9b1b16c7c5ea4cc45800f9bd",
    "id": "2060f98dcfacab5251a93c52d1d1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 3.660081624984741,
    "wer": 0.0,
    "identity": 0.559059202671051,
    "voiceclap": 0.08622735738754272,
    "fitness": 0.8071352442076004,
    "legacy_fitness": 0.8071352442076004,
    "target_cdf": 0.9997931996922089,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.880023002624512,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8518331958588654,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.880023002624512,
        "normalized": 0.9997931996922089,
        "contribution": 0.49989659984610446
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.500368595123291,
        "normalized": 0.41608705873601587,
        "contribution": 0.10402176468400397
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.562960624694824,
        "normalized": 0.9956675509709196,
        "contribution": 0.14935013264563793
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.900958061218262,
        "normalized": 0.9856469868311905,
        "contribution": 0.09856469868311905
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b60b9b433655112b509c559c357010ccd8a03eb81d0b6bc7f509b771ae331bc7",
    "id": "2077dd40f140950874458d2b8b32",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 3,
    "generation": null,
    "ce": 6.499174118041992,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.06668765842914581,
    "fitness": 0.8111392531003668,
    "legacy_fitness": 0.8111392531003668,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8325325187498263,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4842545986175537,
        "normalized": 0.408863767428959,
        "contribution": 0.10221594185723976
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.566638946533203,
        "normalized": 0.9599408341035309,
        "contribution": 0.14399112511552964
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.196876525878906,
        "normalized": 0.9409148467090185,
        "contribution": 0.09409148467090185
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "acaf9b1287fad4f34a876320a27d22dd208c439b803479daadee62781a646b48",
    "id": "2089b16f44b29b59128ca5620736",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": 0.8377411961555481,
    "voiceclap": -0.04298964887857437,
    "fitness": 0.8216393750388022,
    "legacy_fitness": 0.8216393750388022,
    "target_cdf": 0.9915901345965457,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6950535774230957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7552094852049054,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6950535774230957,
        "normalized": 0.9915901345965457,
        "contribution": 0.49579506729827283
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8770133256912231,
        "normalized": 0.17626644875656883,
        "contribution": 0.04406661218914221
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9617295265197754,
        "normalized": 0.8885304554003988,
        "contribution": 0.1332795683100598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4707932472229004,
        "normalized": 0.8206823740743052,
        "contribution": 0.08206823740743052
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4251b8bc021d7b021abdb12e17fb48ef996be8503dc15a374eecfd8cee206ddb",
    "id": "208fb2b27849f68018e57b3fb2f2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04669208824634552,
    "fitness": 0.7948178664942092,
    "legacy_fitness": 0.7948178664942092,
    "target_cdf": 0.9269054842192939,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.72110652923584,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.684890499634573,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.72110652923584,
        "normalized": 0.9269054842192939,
        "contribution": 0.46345274210964693
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4546730518341064,
        "normalized": 0.7803682683335784,
        "contribution": 0.11705524025003676
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0175411701202393,
        "normalized": 0.6969418275486379,
        "contribution": 0.06969418275486379
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4040d688a6097f0eaa1311c2019f4041423e956a3701f3fc8462f521a89eda66",
    "id": "20b726ffd5721e8d46dcbf214f5a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07957921177148819,
    "fitness": 0.8285273881496111,
    "legacy_fitness": 0.8285273881496111,
    "target_cdf": 0.9741313242397945,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.232429027557373,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7042207240298276,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.232429027557373,
        "normalized": 0.9741313242397945,
        "contribution": 0.48706566211989727
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5174179077148438,
        "normalized": 0.0895146120757892,
        "contribution": 0.0223786530189473
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.694857597351074,
        "normalized": 0.8375344060053678,
        "contribution": 0.12563016090080517
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.999939203262329,
        "normalized": 0.6914624799017783,
        "contribution": 0.06914624799017784
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ee7890476a424476f0474c986638fc7a1025a409f3a34ae2c70a9691a1219f2b",
    "id": "20c6aa4d477fe6a9439244a51b52",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.030367851257324,
    "wer": 0.06451612903225806,
    "identity": null,
    "voiceclap": 0.043188031762838364,
    "fitness": 0.7694560767356228,
    "legacy_fitness": 0.7694560767356228,
    "target_cdf": 0.9862070255954634,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5002641677856445,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7511089398056453,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5002641677856445,
        "normalized": 0.9862070255954634,
        "contribution": 0.4931035127977317
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4707932472229004,
        "normalized": 0.8206823740743052,
        "contribution": 0.08206823740743052
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "27daaef294bf757f99b050caff1efb8d76f231e413da32bfff64701adf809b8b",
    "id": "20d7a3b282c93c764309283003ec",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.3504462242126465,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07749072462320328,
    "fitness": 0.8316231718249651,
    "legacy_fitness": 0.8316231718249651,
    "target_cdf": 0.9900472701763179,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.630124092102051,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7363907874812904,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.630124092102051,
        "normalized": 0.9900472701763179,
        "contribution": 0.49502363508815894
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.598835825920105,
        "normalized": 0.10564978306864803,
        "contribution": 0.026412445767162007
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.321176052093506,
        "normalized": 0.7838190079474001,
        "contribution": 0.07838190079474001
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b86660d52c9492010749f4eb3ed9ce4457b08595a4c05332841fb4cbaac0f53a",
    "id": "214a754a0fd7460a743812d1adfc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.061748504638672,
    "wer": 0.0,
    "identity": 0.5116266012191772,
    "voiceclap": -0.06450572609901428,
    "fitness": 0.8221688941187715,
    "legacy_fitness": 0.8221688941187715,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8375960615673028,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.491039514541626,
        "normalized": 0.4119015661257916,
        "contribution": 0.1029753915314479
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.602221965789795,
        "normalized": 0.9625641895690078,
        "contribution": 0.14438462843535116
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.249682903289795,
        "normalized": 0.9462328170567018,
        "contribution": 0.09462328170567019
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "aeebfac3ba64958164f999948f68afeaad0f6dd7361bf2a12f147a353e4c13bf",
    "id": "214c00f9b7b125747204708da037",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.499174118041992,
    "genuineness": 4.2249250411987305,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.24674159288406372,
    "fitness": 0.719841507167286,
    "legacy_fitness": 0.719841507167286,
    "target_cdf": 0.7626472216659115,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.9541230201721191,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5786312915476772,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.9541230201721191,
        "normalized": 0.7626472216659115,
        "contribution": 0.38132361083295574
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8702285289764404,
        "normalized": 0.17425071298327421,
        "contribution": 0.043562678245818554
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.069932699203491,
        "normalized": 0.6684172367937267,
        "contribution": 0.10026258551905899
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5351357460021973,
        "normalized": 0.5348241694984395,
        "contribution": 0.05348241694984395
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d75cec262ad7f309b67f150cdf5c6e25a641351ed88f2dd9d5efc53b2c4af360",
    "id": "2167925120f985c6653c7effb973",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.8546390533447266,
    "wer": 0.0,
    "identity": 0.6029756665229797,
    "voiceclap": 0.0408954918384552,
    "fitness": 0.8044194997891219,
    "legacy_fitness": 0.8044194997891219,
    "target_cdf": 0.9662346008508267,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.110685348510742,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7687113286113059,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.110685348510742,
        "normalized": 0.9662346008508267,
        "contribution": 0.48311730042541334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.124659299850464,
        "normalized": 0.2596155287980314,
        "contribution": 0.06490388219950785
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.07737398147583,
        "normalized": 0.9066346475271524,
        "contribution": 0.13599519712907285
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.589607000350952,
        "normalized": 0.8469494885731179,
        "contribution": 0.0846949488573118
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8a65f8b97b92012706d13f484206bdfa12ca423d78d18a0f59db657a32d88f48",
    "id": "2197ad5376544076a9f013a6469a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.562715530395508,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13046038150787354,
    "fitness": 0.8048159701177751,
    "legacy_fitness": 0.8048159701177751,
    "target_cdf": 0.9247120186784414,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.704874038696289,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6869870603546948,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.704874038696289,
        "normalized": 0.9247120186784414,
        "contribution": 0.4623560093392207
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5525259971618652,
        "normalized": 0.8049331728558216,
        "contribution": 0.12073997592837322
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8899264335632324,
        "normalized": 0.6562844510924838,
        "contribution": 0.06562844510924838
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "271f81edf8ffe550f41e20abcbc50a81a3f0e0f65f5033b76f2dcf684df857c1",
    "id": "21a8f669b985dd1962931c222eb0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.036643981933594,
    "wer": 0.0,
    "identity": 0.5206186175346375,
    "voiceclap": -0.1257377564907074,
    "fitness": 0.739737300575197,
    "legacy_fitness": 0.739737300575197,
    "target_cdf": 0.8217031250405714,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.169203042984009,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5717966119575788,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.169203042984009,
        "normalized": 0.8217031250405714,
        "contribution": 0.4108515625202857
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5716965198516846,
        "normalized": 0.1000529371132764,
        "contribution": 0.0250132342783191
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.7463505268096924,
        "normalized": 0.5601569481462195,
        "contribution": 0.08402354222193292
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4905805587768555,
        "normalized": 0.5190827293704111,
        "contribution": 0.05190827293704112
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0dce5bea1a3fafbee46195c1afdf139e242cba85155837e14d7e942e6e029b98",
    "id": "2215851a2ca48646180fe55275a3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.68307638168335,
    "wer": 0.06666666666666667,
    "identity": 0.4867057204246521,
    "voiceclap": -0.1891106218099594,
    "fitness": 0.7621487135071876,
    "legacy_fitness": 0.7621487135071876,
    "target_cdf": 0.9247120186784414,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.704874038696289,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.699427175924118,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.704874038696289,
        "normalized": 0.9247120186784414,
        "contribution": 0.4623560093392207
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1212668418884277,
        "normalized": 0.25835102183826425,
        "contribution": 0.06458775545956606
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2678627967834473,
        "normalized": 0.7288620844464377,
        "contribution": 0.10932931266696566
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.815117835998535,
        "normalized": 0.6315409845836553,
        "contribution": 0.06315409845836553
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "94e7659d679910da758c7cadf26dc97d625da770095b139be801f4d7f284e46f",
    "id": "228ebde36a09b1d4492746e3a8aa",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07940156012773514,
    "fitness": 0.8163883456549356,
    "legacy_fitness": 0.8163883456549356,
    "target_cdf": 0.9965654549919187,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.019702911376953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8154665606582766,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.019702911376953,
        "normalized": 0.9965654549919187,
        "contribution": 0.49828272749595937
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.226431369781494,
        "normalized": 0.29897739616091534,
        "contribution": 0.07474434904022884
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.135965347290039,
        "normalized": 0.9877755173761718,
        "contribution": 0.14816632760642576
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.214478969573975,
        "normalized": 0.9427315651566268,
        "contribution": 0.09427315651566269
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "440d0ad6ebdb8c1f88413859bacf816012376d31c0e13fd7ed2f4563ad7d56cd",
    "id": "22e74468e000379c919646e91c6a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.545003414154053,
    "wer": 0.06666666666666667,
    "identity": 0.6110373735427856,
    "voiceclap": -0.14667189121246338,
    "fitness": 0.6882700386222306,
    "legacy_fitness": 0.6882700386222306,
    "target_cdf": 0.7826722590503459,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.023110866546631,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6014922676829113,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.023110866546631,
        "normalized": 0.7826722590503459,
        "contribution": 0.39133612952517294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2518744468688965,
        "normalized": 0.3092254511509299,
        "contribution": 0.07730636278773248
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.5384128093719482,
        "normalized": 0.48753504603772674,
        "contribution": 0.07313025690565901
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7139062881469727,
        "normalized": 0.5971951846434683,
        "contribution": 0.05971951846434684
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7b74daedfff83dd0936578fd0000232482ff06e30a6689b1df7a30c8c894f2b9",
    "id": "22e98706c9b1334e7cb95efce8a9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.035824891179800034,
    "fitness": 0.8137247279181192,
    "legacy_fitness": 0.8137247279181192,
    "target_cdf": 0.9755090862302205,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.256777286529541,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8297975482995165,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.256777286529541,
        "normalized": 0.9755090862302205,
        "contribution": 0.48775454311511024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.524963617324829,
        "normalized": 0.42716612713355817,
        "contribution": 0.10679153178338954
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.281975746154785,
        "normalized": 0.9331927987311419,
        "contribution": 0.13997891980967128
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.320091247558594,
        "normalized": 0.9527255359134541,
        "contribution": 0.09527255359134541
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "00e45e856a59b5db3513531745c1945ee2de191bcfcf98706445ce5a67b61c97",
    "id": "2300ef467641f97b3fb0f7f418dd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.432034492492676,
    "wer": 0.0,
    "identity": 0.5220077633857727,
    "voiceclap": -0.17046210169792175,
    "fitness": 0.8197230234467276,
    "legacy_fitness": 0.8197230234467276,
    "target_cdf": 0.9534909436550316,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.956477165222168,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7058425652226269,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.956477165222168,
        "normalized": 0.9534909436550316,
        "contribution": 0.4767454718275158
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8837981224060059,
        "normalized": 0.17829687836447006,
        "contribution": 0.044574219591117514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.525838851928711,
        "normalized": 0.7984049796580064,
        "contribution": 0.11976074694870095
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.863523483276367,
        "normalized": 0.6476212685529261,
        "contribution": 0.06476212685529262
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "71274c83de84b57f9e619e67c775baf509fdbda273fe69f6829171219a60c9c1",
    "id": "23795044859d04f78e5d05bcf687",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.062326908111572,
    "genuineness": 4.193545341491699,
    "wer": 0.0,
    "identity": 0.5092217326164246,
    "voiceclap": 0.024984534829854965,
    "fitness": 0.8045160632206088,
    "legacy_fitness": 0.8045160632206088,
    "target_cdf": 0.956453669244664,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9889419078826904,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7967710024802037,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9889419078826904,
        "normalized": 0.956453669244664,
        "contribution": 0.478226834622332
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.452026844024658,
        "normalized": 0.3945120902788001,
        "contribution": 0.09862802256970002
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9617295265197754,
        "normalized": 0.8885304554003988,
        "contribution": 0.1332795683100598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.686418056488037,
        "normalized": 0.8663657697811189,
        "contribution": 0.08663657697811189
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8867d9dcf2475bda4d636023916dce460c56232770e5f419e83ccd6885b51c75",
    "id": "23fa40e56bea763a89e51035dcd6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": 0.6026482582092285,
    "voiceclap": -0.058837685734033585,
    "fitness": 0.8295469551970394,
    "legacy_fitness": 0.8295469551970394,
    "target_cdf": 0.9940589248517815,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.824913501739502,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7886655743081585,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.824913501739502,
        "normalized": 0.9940589248517815,
        "contribution": 0.49702946242589074
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0602035522460938,
        "normalized": 0.23614752264221534,
        "contribution": 0.059036880660553834
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.39762020111084,
        "normalized": 0.9453738037624609,
        "contribution": 0.14180607056436914
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9328465461730957,
        "normalized": 0.9079316065734483,
        "contribution": 0.09079316065734483
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2132f5bb4399db71540c39b6e03f68eb60ec4d6984a2ff8db6db76bb5181a91f",
    "id": "2688c2cee36f8ec2535d182b44dd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": 0.7647292613983154,
    "voiceclap": 0.05629081279039383,
    "fitness": 0.8214061504006502,
    "legacy_fitness": 0.8214061504006502,
    "target_cdf": 0.999200941205048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.490444183349609,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7952742191980574,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.490444183349609,
        "normalized": 0.999200941205048,
        "contribution": 0.499600470602524
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.033064126968384,
        "normalized": 0.2266273368759334,
        "contribution": 0.05665683421898335
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.8157196044921875,
        "normalized": 0.9755091035016348,
        "contribution": 0.14632636552524522
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.073662757873535,
        "normalized": 0.926905488513048,
        "contribution": 0.0926905488513048
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "21cd5ddd589293de0eee6cbde35b8a793829f986cdf19a660ce44c61feb40349",
    "id": "26bf5a527293c2840569c7fcf974",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04320230707526207,
    "fitness": 0.8218463422716711,
    "legacy_fitness": 0.8218463422716711,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7366616096233841,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5784813165664673,
        "normalized": 0.10143141696818753,
        "contribution": 0.02535785424204688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.41798734664917,
        "normalized": 0.8081485455599589,
        "contribution": 0.0808148545559959
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "068580282d2947c7c27e796cc64b9df193025506db0178e67c81eaac599b21e0",
    "id": "26e7b4077993141639d9b7f88010",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.515059471130371,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": 0.5996673107147217,
    "voiceclap": 0.02719230204820633,
    "fitness": 0.7452902780730065,
    "legacy_fitness": 0.7452902780730065,
    "target_cdf": 0.8237335655606859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.177319288253784,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6360179962807043,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.177319288253784,
        "normalized": 0.8237335655606859,
        "contribution": 0.41186678278034294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1110897064208984,
        "normalized": 0.25457682692833283,
        "contribution": 0.06364420673208321
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8286361694335938,
        "normalized": 0.5884785213525642,
        "contribution": 0.08827177820288462
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1011507511138916,
        "normalized": 0.7223522856539355,
        "contribution": 0.07223522856539355
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0d120f3619a6764b74858d13e516df4986a0d426712523615a3f28aa0da83aa8",
    "id": "26e8dc2150c75529a1b64d0f5063",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.260893821716309,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.014999051578342915,
    "fitness": 0.8379368759223155,
    "legacy_fitness": 0.8379368759223155,
    "target_cdf": 0.988262823627061,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5651941299438477,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7755867225799795,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5651941299438477,
        "normalized": 0.988262823627061,
        "contribution": 0.4941314118135305
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.066988229751587,
        "normalized": 0.2385614708758641,
        "contribution": 0.059640367718966025
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6160099506378174,
        "normalized": 0.852421372162536,
        "contribution": 0.08524213721625361
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "73ebfc47d5233e9a45fb8306313179d90ee0ae0950acc3e909b8baf44eed2f5c",
    "id": "26f9bd24036880fae8fe2316bf7a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.260893821716309,
    "genuineness": 4.055472373962402,
    "wer": 0.0,
    "identity": 0.5713537931442261,
    "voiceclap": 0.030083253979682922,
    "fitness": 0.8251604989755578,
    "legacy_fitness": 0.8251604989755578,
    "target_cdf": 0.9993202757017665,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.539141654968262,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7664752541032288,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.539141654968262,
        "normalized": 0.9993202757017665,
        "contribution": 0.49966013785088326
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.707392930984497,
        "normalized": 0.130294529860791,
        "contribution": 0.03257363246519775
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.602221965789795,
        "normalized": 0.9625641895690078,
        "contribution": 0.14438462843535116
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.871239185333252,
        "normalized": 0.8985685535179664,
        "contribution": 0.08985685535179665
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1fffa0caef8886b4eceeb7c23e4a59668526e59debac2192b40edbfeb49d6a62",
    "id": "27219cfc10ce9e85483a838a9661",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 5.935244083404541,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.031538914889097214,
    "fitness": 0.807124593738235,
    "legacy_fitness": 0.807124593738235,
    "target_cdf": 0.9763929662774343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.273009777069092,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8259659074223583,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.273009777069092,
        "normalized": 0.9763929662774343,
        "contribution": 0.48819648313871716
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4689886569976807,
        "normalized": 0.40204902050850394,
        "contribution": 0.10051225512712599
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.415411949157715,
        "normalized": 0.9470811189066717,
        "contribution": 0.14206216783600076
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.3112897872924805,
        "normalized": 0.9519500132051444,
        "contribution": 0.09519500132051445
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4463d5a54cada41df846c9621036578a8d1c905988137eb90d7b46474779366a",
    "id": "273aeb414dcc733602950ed8e7ad",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09425752609968185,
    "fitness": 0.797742812162175,
    "legacy_fitness": 0.797742812162175,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6666420970996991,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.039849042892456,
        "normalized": 0.2289868888511522,
        "contribution": 0.05724672221278805
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.900913953781128,
        "normalized": 0.6129857446086481,
        "contribution": 0.0919478616912972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.645698308944702,
        "normalized": 0.5735999100129361,
        "contribution": 0.05735999100129361
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "730f5d7a7623a975fda7db4669f0dffbd1e97af2a2f5a9a7d9accb565a6fd3cc",
    "id": "2752f60fdfc011554db63e054ab5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.133810997009277,
    "genuineness": 3.9676074981689453,
    "wer": 0.06666666666666667,
    "identity": 0.48371291160583496,
    "voiceclap": 0.05457556992769241,
    "fitness": 0.7604819484517961,
    "legacy_fitness": 0.7604819484517961,
    "target_cdf": 0.9922796672226393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7275185585021973,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8276431492414585,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7275185585021973,
        "normalized": 0.9922796672226393,
        "contribution": 0.49613983361131964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4282798767089844,
        "normalized": 0.38402678994719125,
        "contribution": 0.09600669748679781
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.495473384857178,
        "normalized": 0.9542463876094935,
        "contribution": 0.14313695814142402
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.047259330749512,
        "normalized": 0.9235966000191711,
        "contribution": 0.09235966000191712
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "734ad0851fc3cd0dcca3a8a4fa9a250c12facd89a5e7461b68cfb4418ebdfae2",
    "id": "282c3cde8dc34869381a0908cb59",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.835810661315918,
    "wer": 0.0,
    "identity": 0.7142549157142639,
    "voiceclap": -0.053777072578668594,
    "fitness": 0.7639323951880141,
    "legacy_fitness": 0.7639323951880141,
    "target_cdf": 0.8900064710934966,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4857358932495117,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6576870930163582,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4857358932495117,
        "normalized": 0.8900064710934966,
        "contribution": 0.4450032355467483
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3612678050994873,
        "normalized": 0.7553448566210981,
        "contribution": 0.11330172849316471
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8745245933532715,
        "normalized": 0.6512405138489399,
        "contribution": 0.065124051384894
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d54073a6a80f60af02b17535bd1e1da6318fa4693ac229a78a7905c59987fe9b",
    "id": "299c732c7615ad935c4be20ba135",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.074300289154053,
    "wer": 0.0,
    "identity": 0.6061183214187622,
    "voiceclap": -0.024927668273448944,
    "fitness": 0.787146371805467,
    "legacy_fitness": 0.787146371805467,
    "target_cdf": 0.912986084230339,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.623711585998535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7004997898128482,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.623711585998535,
        "normalized": 0.912986084230339,
        "contribution": 0.4564930421151695
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9550387859344482,
        "normalized": 0.20049800895093267,
        "contribution": 0.05012450223773317
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.494703769683838,
        "normalized": 0.7906255737437923,
        "contribution": 0.11859383606156884
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2067627906799316,
        "normalized": 0.7528840939837675,
        "contribution": 0.07528840939837676
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f59e9e2353c5a522e24797743b3cdb5f7867ed6e6cbb615de9390a114f1ec9ef",
    "id": "29ebaeac0efcf1301b95590f2490",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.835810661315918,
    "wer": 0.0,
    "identity": 0.6529397964477539,
    "voiceclap": 0.08193185925483704,
    "fitness": 0.814156421121333,
    "legacy_fitness": 0.814156421121333,
    "target_cdf": 0.9904545229601343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6463565826416016,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7393521532655913,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6463565826416016,
        "normalized": 0.9904545229601343,
        "contribution": 0.4952272614800671
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5377724170684814,
        "normalized": 0.09336535255322709,
        "contribution": 0.023341338138306772
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.362037181854248,
        "normalized": 0.9418287736823874,
        "contribution": 0.1412743160523581
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3651812076568604,
        "normalized": 0.7950923759485932,
        "contribution": 0.07950923759485932
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6e7947b29998576de0586147819e5148cae2478b5b5334f9fcfb50e3ae44276a",
    "id": "29fff870a89f38b69d31b412e58c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": 0.5979686975479126,
    "voiceclap": -0.04106009006500244,
    "fitness": 0.7679347989130306,
    "legacy_fitness": 0.7679347989130306,
    "target_cdf": 0.8943502420521354,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.510084629058838,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6680664623858803,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.510084629058838,
        "normalized": 0.8943502420521354,
        "contribution": 0.4471751210260677
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6147961616516113,
        "normalized": 0.8196580135026951,
        "contribution": 0.12294870202540426
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9691355228424072,
        "normalized": 0.6817707531818977,
        "contribution": 0.06817707531818977
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c3f1c6b0257994d4d9b955a601cc27ee8b505bfd7012565832f1358f9662b15d",
    "id": "2a4ac7728d3b727291af95cd327b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.519899368286133,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13998223841190338,
    "fitness": 0.8287642083082091,
    "legacy_fitness": 0.8287642083082091,
    "target_cdf": 0.9650455080496254,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0944528579711914,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7286254701102208,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0944528579711914,
        "normalized": 0.9650455080496254,
        "contribution": 0.4825227540248127
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9923553466796875,
        "normalized": 0.2127618447245091,
        "contribution": 0.053190461181127274
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5436301231384277,
        "normalized": 0.8027714861707109,
        "contribution": 0.12041572292560664
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1099517345428467,
        "normalized": 0.7249653197867423,
        "contribution": 0.07249653197867423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8db8cb38061b0b29c7a3b5d0de0b5fa95de370f60012d7ccc2e4e27664e03bac",
    "id": "2a5678be4d51cc0ee8f5ecbd22f7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.049196243286133,
    "wer": 0.0,
    "identity": 0.6482986807823181,
    "voiceclap": -0.08279784768819809,
    "fitness": 0.7151138942948393,
    "legacy_fitness": 0.7151138942948393,
    "target_cdf": 0.7710131330314103,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.9825297594070435,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.582760506738873,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.9825297594070435,
        "normalized": 0.7710131330314103,
        "contribution": 0.38550656651570514
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.995747685432434,
        "normalized": 0.2138980203686049,
        "contribution": 0.05347450509215122
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.807508707046509,
        "normalized": 0.5812444562175472,
        "contribution": 0.08718666843263208
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6236958503723145,
        "normalized": 0.5659276669838458,
        "contribution": 0.05659276669838458
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4999702a29cd2dd9fe9331bb4b8512e4c8ca858b3adb014ffb9f09a899cfe918",
    "id": "2a765726016b043a82cc1350f781",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.546830177307129,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.046579405665397644,
    "fitness": 0.818083694895716,
    "legacy_fitness": 0.818083694895716,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7342196992337415,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5513420104980469,
        "normalized": 0.09599963185428989,
        "contribution": 0.023999907963572473
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.228601455688477,
        "normalized": 0.9269054845040974,
        "contribution": 0.1390358226756146
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3431787490844727,
        "normalized": 0.7895000148839948,
        "contribution": 0.07895000148839948
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "085637a16a4a4c20ddbc337ee62f7a2769a13b3c2b4fe296a4ee5777df27ef50",
    "id": "2abf9e8d4f53017f49e92e372672",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.3127899169921875,
    "wer": 0.0,
    "identity": 0.6719297170639038,
    "voiceclap": -0.1323479413986206,
    "fitness": 0.749742267142399,
    "legacy_fitness": 0.749742267142399,
    "target_cdf": 0.8155238263024167,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.1448543071746826,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6143009712008048,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.1448543071746826,
        "normalized": 0.8155238263024167,
        "contribution": 0.40776191315120836
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.887190580368042,
        "normalized": 0.17931761586218503,
        "contribution": 0.04482940396554626
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0299019813537598,
        "normalized": 0.6555656036609419,
        "contribution": 0.09833484054914128
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.821718692779541,
        "normalized": 0.6337481353490892,
        "contribution": 0.06337481353490892
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0b485f607f09b79198bf2e5d6d513767cee04f07506f0b6adb52ff68088fc42e",
    "id": "2b937f572be31eaa2562f9b0c3bd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 4.306513786315918,
    "wer": 0.06666666666666667,
    "identity": 0.4296848773956299,
    "voiceclap": -0.010638609528541565,
    "fitness": 0.7727635102759277,
    "legacy_fitness": 0.7727635102759277,
    "target_cdf": 0.9832067001342365,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.419102191925049,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7640553928979361,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.419102191925049,
        "normalized": 0.9832067001342365,
        "contribution": 0.49160335006711825
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9889628887176514,
        "normalized": 0.21162916044311025,
        "contribution": 0.05290729011077756
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.059582710266113,
        "normalized": 0.9040003875053659,
        "contribution": 0.1356000581258049
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5544028282165527,
        "normalized": 0.8394469459423535,
        "contribution": 0.08394469459423536
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e6009bf4f33bcc8db8995f1d44556348bb9ef82ffdb23951e3162241e0fa5a9f",
    "id": "2cddb0dae181f220e6cd483767e3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": 0.6576404571533203,
    "voiceclap": 0.007867230102419853,
    "fitness": 0.7613434521900547,
    "legacy_fitness": 0.7613434521900547,
    "target_cdf": 0.8855360207054879,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4613871574401855,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6819667625350097,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4613871574401855,
        "normalized": 0.8855360207054879,
        "contribution": 0.44276801035274393
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8939753770828247,
        "normalized": 0.18137005253541438,
        "contribution": 0.045342513133853596
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.588109016418457,
        "normalized": 0.8134348350041303,
        "contribution": 0.12201522525061953
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.087949275970459,
        "normalized": 0.7184101379779255,
        "contribution": 0.07184101379779255
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4686ef7bfa656419e2257f2b1b3ac74cc6cea426d262bfa6c5b6d53f853e7807",
    "id": "2d327ee6ec09aa936deaad302159",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.469690799713135,
    "wer": 0.0,
    "identity": 0.5487720966339111,
    "voiceclap": -0.23818056285381317,
    "fitness": 0.777215074360595,
    "legacy_fitness": 0.777215074360595,
    "target_cdf": 0.8697054919803655,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3802249431610107,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6026804168035631,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3802249431610107,
        "normalized": 0.8697054919803655,
        "contribution": 0.43485274599018275
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6666839122772217,
        "normalized": 0.12062361006755057,
        "contribution": 0.030155902516887642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.920929193496704,
        "normalized": 0.6196978959173229,
        "contribution": 0.09295468438759844
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.2870571613311768,
        "normalized": 0.4471708390889426,
        "contribution": 0.04471708390889426
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1ab34bb4111bef85e61931b4a5cf30ee963263e1f50a5f5a0c6aa5cae7b243e0",
    "id": "2d4651385345b6405957ecf93930",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.586543083190918,
    "genuineness": 4.6203155517578125,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13292399048805237,
    "fitness": 0.7612509764267175,
    "legacy_fitness": 0.7612509764267175,
    "target_cdf": 0.8081485027140713,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.1164474487304688,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5886804947890721,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.1164474487304688,
        "normalized": 0.8081485027140713,
        "contribution": 0.4040742513570357
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8464816808700562,
        "normalized": 0.16731154547607663,
        "contribution": 0.04182788636901916
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.807508707046509,
        "normalized": 0.5812444562175472,
        "contribution": 0.08718666843263208
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.595092535018921,
        "normalized": 0.5559168863038523,
        "contribution": 0.05559168863038523
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a2d58c572dc4c45ade092071f68a92708f44353436a0646da5d02c26e5adccdd",
    "id": "2d4b2920ed5a31d18fb5f4d48320",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.532451152801514,
    "wer": 0.03571428571428571,
    "identity": 0.6110221743583679,
    "voiceclap": -0.09015659242868423,
    "fitness": 0.8119692420274185,
    "legacy_fitness": 0.8119692420274185,
    "target_cdf": 0.9877755264344382,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.548961639404297,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.741780962113329,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.548961639404297,
        "normalized": 0.9877755264344382,
        "contribution": 0.4938877632172191
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8396968841552734,
        "normalized": 0.16536210020893177,
        "contribution": 0.04134052505223294
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9617295265197754,
        "normalized": 0.8885304554003988,
        "contribution": 0.1332795683100598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.136354923248291,
        "normalized": 0.7327310553381711,
        "contribution": 0.0732731055338171
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "63aa8974e88a58d761bf5629804b8aa8a94c304dd5cf16bf55363e903dd55e15",
    "id": "2d819fa0c2a8ba65c2be11029f1f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": 0.7441778182983398,
    "voiceclap": -0.03297567367553711,
    "fitness": 0.8294894819473693,
    "legacy_fitness": 0.8294894819473693,
    "target_cdf": 0.9872707933218218,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.532729148864746,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7353136588879738,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.532729148864746,
        "normalized": 0.9872707933218218,
        "contribution": 0.4936353966609109
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7571277618408203,
        "normalized": 0.8506121673397089,
        "contribution": 0.12759182510095632
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.360780715942383,
        "normalized": 0.7939810260608104,
        "contribution": 0.07939810260608104
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e9007fb090e79220b277dafebd3c9cadd60d910116fa1048591fa612136e976e",
    "id": "2d848519d16c0aadbe8820fad64d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.229123115539551,
    "genuineness": 4.582659721374512,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.17726640403270721,
    "fitness": 0.7518092008637391,
    "legacy_fitness": 0.7518092008637391,
    "target_cdf": 0.8102739358359581,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.124563694000244,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5968742056308176,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.124563694000244,
        "normalized": 0.8102739358359581,
        "contribution": 0.40513696791797904
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9855704307556152,
        "normalized": 0.2105000137670302,
        "contribution": 0.05262500344175755
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.840867757797241,
        "normalized": 0.5926530902100725,
        "contribution": 0.08889796353151087
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4427249431610107,
        "normalized": 0.5021427073957011,
        "contribution": 0.05021427073957011
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "373fdc57ba154444cdf992a0f2e7a9f004a9d7d8d04fd3dee49cc2dad4834f69",
    "id": "2d885e338b22b5b422f42d396bb4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.143336772918701,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.007605614140629768,
    "fitness": 0.8025973111225915,
    "legacy_fitness": 0.8025973111225915,
    "target_cdf": 0.9409148593413338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.834733486175537,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7414384260907885,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.834733486175537,
        "normalized": 0.9409148593413338,
        "contribution": 0.4704574296706669
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1212668418884277,
        "normalized": 0.25835102183826425,
        "contribution": 0.06458775545956606
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6770663261413574,
        "normalized": 0.8336650598974741,
        "contribution": 0.12504975898462112
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4399898052215576,
        "normalized": 0.8134348197593442,
        "contribution": 0.08134348197593443
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c7d9feca8159323e11d9f5225f64ad412cf1ad34cb6bb7c713541cfa28fe03c6",
    "id": "2d8ec5ee2a03563ae8d7d3393840",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": 0.8057564496994019,
    "voiceclap": -0.02887561172246933,
    "fitness": 0.7677362563876191,
    "legacy_fitness": 0.7677362563876191,
    "target_cdf": 0.8793763603106894,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.428922176361084,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6361842564778941,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.428922176361084,
        "normalized": 0.8793763603106894,
        "contribution": 0.4396881801553447
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6463295221328735,
        "normalized": 0.11598248129215816,
        "contribution": 0.02899562032303954
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.285654067993164,
        "normalized": 0.7340144537731833,
        "contribution": 0.11010216806597749
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.646798610687256,
        "normalized": 0.573982879335324,
        "contribution": 0.0573982879335324
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e516079a3336ac2e64c66d02747ae31a638d72e899373b81afe1bbd563eba7a9",
    "id": "2d93f9965396fcfefc7398523423",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.499174118041992,
    "genuineness": 4.124508857727051,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0917443186044693,
    "fitness": 0.7955682533389272,
    "legacy_fitness": 0.7955682533389272,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6663126264979896,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8770133256912231,
        "normalized": 0.17626644875656883,
        "contribution": 0.04406661218914221
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2144882678985596,
        "normalized": 0.7131122892698544,
        "contribution": 0.10696684339047816
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5510876178741455,
        "normalized": 0.540448062691189,
        "contribution": 0.054044806269118895
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "39a17e9f85f79dc4e1a404910ff66a3ec0b7e06d7926331f5b470810e07f50ca",
    "id": "2de5e591589339a8d3df46bab4fe",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": 0.44415321946144104,
    "voiceclap": -0.002604275941848755,
    "fitness": 0.8426401279433479,
    "legacy_fitness": 0.8426401279433479,
    "target_cdf": 0.9968750600147067,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.052167892456055,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7349949421949574,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.052167892456055,
        "normalized": 0.9968750600147067,
        "contribution": 0.49843753000735336
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5038483142852783,
        "normalized": 0.08701389328623654,
        "contribution": 0.021753473321559136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.255288600921631,
        "normalized": 0.9301035297844544,
        "contribution": 0.13951552946766815
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2067627906799316,
        "normalized": 0.7528840939837675,
        "contribution": 0.07528840939837676
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9207e906502a181e182c6aaae9642f7f988e3ba961d2b552ee92fff2946d1b14",
    "id": "2deb6e5b862e43d0b6925a6f7365",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.16,
    "cfg": 3,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0766865462064743,
    "fitness": 0.8342758434261653,
    "legacy_fitness": 0.8342758434261653,
    "target_cdf": 0.996049248624995,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.971005439758301,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8190433102326735,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.971005439758301,
        "normalized": 0.996049248624995,
        "contribution": 0.4980246243124975
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.372305154800415,
        "normalized": 0.35965731004855184,
        "contribution": 0.08991432751213796
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.317558765411377,
        "normalized": 0.9371462024575616,
        "contribution": 0.14057193036863425
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9152445793151855,
        "normalized": 0.9053242803940382,
        "contribution": 0.09053242803940383
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "59ac93d5f1270d5de4bec3292cb98439e70856a85d545bb245d33c7eb752443a",
    "id": "2e086e391c3aefa94889e1d2762a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 3.6663577556610107,
    "wer": 0.0,
    "identity": 0.7214545011520386,
    "voiceclap": 0.056964997202157974,
    "fitness": 0.8054925574070328,
    "legacy_fitness": 0.8054925574070328,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7351520133522794,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5309876203536987,
        "normalized": 0.09206841268091792,
        "contribution": 0.02301710317022948
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.255288600921631,
        "normalized": 0.9301035297844544,
        "contribution": 0.13951552946766815
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4003851413726807,
        "normalized": 0.803854136082268,
        "contribution": 0.0803854136082268
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "41b221fab65703a88806c28e4c89c8368aeab8d589bf2e0cff4a66543acf6c68",
    "id": "2e5f52bb79b38bc4627fc28f3f84",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 3.8295347690582275,
    "wer": 0.0,
    "identity": 0.564170777797699,
    "voiceclap": 0.01641111634671688,
    "fitness": 0.808070389875486,
    "legacy_fitness": 0.808070389875486,
    "target_cdf": 0.9867481280341037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5164966583251953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7379974555222508,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5164966583251953,
        "normalized": 0.9867481280341037,
        "contribution": 0.49337406401705186
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5174179077148438,
        "normalized": 0.0895146120757892,
        "contribution": 0.0223786530189473
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.466392755508423,
        "normalized": 0.8196579878328937,
        "contribution": 0.08196579878328937
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "85f695486e7587216ee34574c57b1a770b56dff9800bb6e126e9254dbe9e4387",
    "id": "2ec1a069bd624e502ea779933ee2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": 0.4799972474575043,
    "voiceclap": -0.10081747174263,
    "fitness": 0.8336591901361636,
    "legacy_fitness": 0.8336591901361636,
    "target_cdf": 0.9967235449909552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.035935401916504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7400395732252243,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.035935401916504,
        "normalized": 0.9967235449909552,
        "contribution": 0.4983617724954776
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.632759928703308,
        "normalized": 0.11295969157872005,
        "contribution": 0.02823992289468001
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2111635208129883,
        "normalized": 0.754116172534443,
        "contribution": 0.07541161725344431
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8225a1d693070c327a52774d9c67859d85ce6897b9649fc842ebc07d428f3fc6",
    "id": "2eff5db2eaeb6524f3fad0ecf678",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": 0.7842475175857544,
    "voiceclap": 0.08408468961715698,
    "fitness": 0.8033458648066446,
    "legacy_fitness": 0.8033458648066446,
    "target_cdf": 0.9445039705317666,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8671984672546387,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7099353508745253,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8671984672546387,
        "normalized": 0.9445039705317666,
        "contribution": 0.4722519852658833
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8329119682312012,
        "normalized": 0.1634273784096072,
        "contribution": 0.0408568446024018
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6592745780944824,
        "normalized": 0.8297365948158377,
        "contribution": 0.12446048922237565
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.105551242828369,
        "normalized": 0.7236603178386457,
        "contribution": 0.07236603178386457
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "37329c743b2f0d8eeaeefaf796cc4ac1a9c20cab645f2d7e732c3e7cc952a984",
    "id": "2f51c709b7dd48a51a42da24b348",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.0305562019348145,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.006774710491299629,
    "fitness": 0.82782859085007,
    "legacy_fitness": 0.82782859085007,
    "target_cdf": 0.9952532227437393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9060754776000977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7896321947303144,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9060754776000977,
        "normalized": 0.9952532227437393,
        "contribution": 0.49762661137186964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0262794494628906,
        "normalized": 0.2242816522895154,
        "contribution": 0.05607041307237885
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.5310564041137695,
        "normalized": 0.957170068560992,
        "contribution": 0.14357551028414878
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.047259330749512,
        "normalized": 0.9235966000191711,
        "contribution": 0.09235966000191712
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "84a3eca9a12a57c750b8d3b5fd07925e24dba9d4fbb4acfe22591e524a6e82dc",
    "id": "2f59aaadeac4a14cfe8cc5797d28",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.054384231567383,
    "genuineness": 4.111956596374512,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.05031442269682884,
    "fitness": 0.7559905993639524,
    "legacy_fitness": 0.7559905993639524,
    "target_cdf": 0.9745975945036527,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2405447959899902,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7728545894557534,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2405447959899902,
        "normalized": 0.9745975945036527,
        "contribution": 0.48729879725182634
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.073773145675659,
        "normalized": 0.24098894914960267,
        "contribution": 0.06024723728740067
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.246392726898193,
        "normalized": 0.9290496824330181,
        "contribution": 0.1393574523649527
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.651214122772217,
        "normalized": 0.8595110255157367,
        "contribution": 0.08595110255157368
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7118fe7927e0f1b9a368d9aa7f3b8ee6681c9e9f36f26d07752291f789d3f084",
    "id": "2f812ba73055f9b7a1fcd11ccf03",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.8546390533447266,
    "wer": 0.0,
    "identity": 0.6550590991973877,
    "voiceclap": -0.04653815180063248,
    "fitness": 0.7784112297748866,
    "legacy_fitness": 0.7784112297748866,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7660733953467201,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3282036781311035,
        "normalized": 0.3408456226113612,
        "contribution": 0.0852114056528403
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9261465072631836,
        "normalized": 0.8824847562300842,
        "contribution": 0.13237271343451262
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.783229351043701,
        "normalized": 0.8840175406504699,
        "contribution": 0.088401754065047
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9ff3a0496430d767d1073831bd6d0917ba6ab105ff6cf08ac449a94544347aa4",
    "id": "2fbd17a6601e7639ba6bb3f75128",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09152843058109283,
    "fitness": 0.8246942194653524,
    "legacy_fitness": 0.8246942194653524,
    "target_cdf": 0.9711807712620442,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1837315559387207,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6926160238763495,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1837315559387207,
        "normalized": 0.9711807712620442,
        "contribution": 0.4855903856310221
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4292153120040894,
        "normalized": 0.0741850635010084,
        "contribution": 0.0185462658752521
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6414833068847656,
        "normalized": 0.8257492948479319,
        "contribution": 0.12386239422718978
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8591229915618896,
        "normalized": 0.6461697814288547,
        "contribution": 0.06461697814288547
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3945e53acf98b30fc4b69e6e5b23c34a53f49834b97efa2598db771b5b9e24ac",
    "id": "2fcb6923086cbbc024c03b96b735",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04969632998108864,
    "fitness": 0.7722447060215353,
    "legacy_fitness": 0.7722447060215353,
    "target_cdf": 0.8793763603106894,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.428922176361084,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.675629761347239,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.428922176361084,
        "normalized": 0.8793763603106894,
        "contribution": 0.4396881801553447
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9346842765808105,
        "normalized": 0.19399141727969682,
        "contribution": 0.048497854319924205
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4413294792175293,
        "normalized": 0.7768860847795082,
        "contribution": 0.11653291271692622
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.057145595550537,
        "normalized": 0.7091081415504378,
        "contribution": 0.07091081415504379
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "faae4868cb91129f96fb45b580bffdf474de2328f3452b243e3ed42fa716df5a",
    "id": "2fece1c8681a8fd920449cdd174b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04532899707555771,
    "fitness": 0.8230734293674895,
    "legacy_fitness": 0.8230734293674895,
    "target_cdf": 0.9872707933218218,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.532729148864746,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7472891850781236,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.532729148864746,
        "normalized": 0.9872707933218218,
        "contribution": 0.4936353966609109
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.707392930984497,
        "normalized": 0.130294529860791,
        "contribution": 0.03257363246519775
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.210809707641602,
        "normalized": 0.9247119875768577,
        "contribution": 0.13870679813652864
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.483994960784912,
        "normalized": 0.8237335781548634,
        "contribution": 0.08237335781548634
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c68b246a2b57e797f9b37653c900191f08888a94b8a4384379ec24cdeedc6f73",
    "id": "2ff4e6cf5d90eb6f4900fef7ca62",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.55477237701416,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0329832062125206,
    "fitness": 0.781597514739012,
    "legacy_fitness": 0.781597514739012,
    "target_cdf": 0.8870403206157192,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.469503402709961,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6302544556043456,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.469503402709961,
        "normalized": 0.8870403206157192,
        "contribution": 0.4435201603078596
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.112187385559082,
        "normalized": 0.6817707678889575,
        "contribution": 0.10226561518334362
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5697896480560303,
        "normalized": 0.5470311609692374,
        "contribution": 0.05470311609692374
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f2efbaf918e38961545928ee105eba5af3ebf88d7010dce6fec60c0c1d78cdeb",
    "id": "303982712df7f143270be2d7aaae",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.157639026641846,
    "genuineness": 4.432034492492676,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10256016999483109,
    "fitness": 0.7399948581981602,
    "legacy_fitness": 0.7399948581981602,
    "target_cdf": 0.8027715573565053,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.0961570739746094,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6313646990048329,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.0961570739746094,
        "normalized": 0.8027715573565053,
        "contribution": 0.40138577867825265
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.131443977355957,
        "normalized": 0.2621540050393544,
        "contribution": 0.0655385012598386
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.9164814949035645,
        "normalized": 0.6182093832757598,
        "contribution": 0.09273140749136397
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0835487842559814,
        "normalized": 0.7170901157537775,
        "contribution": 0.07170901157537775
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "374bf565811a662d2f2696c8062c04bb20e7309e747ef6f92aad99017d2b2bd5",
    "id": "3106492b2b536fe3333e021e90f4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.18857964873313904,
    "fitness": 0.8320243335791758,
    "legacy_fitness": 0.8320243335791758,
    "target_cdf": 0.988262823627061,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5651941299438477,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7502466230049538,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5651941299438477,
        "normalized": 0.988262823627061,
        "contribution": 0.4941314118135305
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.326454162597656,
        "normalized": 0.9381054562342774,
        "contribution": 0.1407158184351416
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.395984649658203,
        "normalized": 0.8027715080715057,
        "contribution": 0.08027715080715057
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "751f189e4c4f98458baf0f111c080b490be189ce64c3fd53ecd6742d6e597aaa",
    "id": "31303805a3946831f3097e8a718f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.004093565046787262,
    "fitness": 0.8252099773702175,
    "legacy_fitness": 0.8252099773702175,
    "target_cdf": 0.9706627967789073,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1756153106689453,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7027666771446834,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1756153106689453,
        "normalized": 0.9706627967789073,
        "contribution": 0.48533139838945366
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5581269264221191,
        "normalized": 0.09733711108070414,
        "contribution": 0.024334277770176035
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7393362522125244,
        "normalized": 0.8469494732316728,
        "contribution": 0.1270424209847509
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.903127908706665,
        "normalized": 0.6605858000030286,
        "contribution": 0.06605858000030286
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "80b50e402484c0d29bd90b515acfbd05d62dbd1eb77aa3dfe30eb1b27326256f",
    "id": "314817692a4ce4b553ab8d3e20c0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.086154937744141,
    "genuineness": 4.20609712600708,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09671874344348907,
    "fitness": 0.826829835497789,
    "legacy_fitness": 0.826829835497789,
    "target_cdf": 0.9988438302744477,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.376816749572754,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7701362237120641,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.376816749572754,
        "normalized": 0.9988438302744477,
        "contribution": 0.49942191513722384
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7752411365509033,
        "normalized": 0.1475786358230735,
        "contribution": 0.03689465895576838
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.602221965789795,
        "normalized": 0.9625641895690078,
        "contribution": 0.14438462843535116
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8448362350463867,
        "normalized": 0.8943502118372078,
        "contribution": 0.08943502118372078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b3eb10071a093de14fb701269797a541567bde0122cc575ed31927ee7379199c",
    "id": "32a1bc6ec4585d9470f15fb97b6a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": 0.6202124953269958,
    "voiceclap": -0.07859587669372559,
    "fitness": 0.8249723574933558,
    "legacy_fitness": 0.8249723574933558,
    "target_cdf": 0.9706627967789073,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1756153106689453,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7097881025380195,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1756153106689453,
        "normalized": 0.9706627967789073,
        "contribution": 0.48533139838945366
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5970046520233154,
        "normalized": 0.8155238285113872,
        "contribution": 0.12232857427670807
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0263421535491943,
        "normalized": 0.6996650747455826,
        "contribution": 0.06996650747455826
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a15e34995c95e7f60b7834b902ddc972031789811df0b1623acbebeb5519bffa",
    "id": "32a4aa5ba844c58ceb7622eec64c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.9425034523010254,
    "wer": 0.0,
    "identity": 0.7038818597793579,
    "voiceclap": 0.06261125206947327,
    "fitness": 0.8184538453768972,
    "legacy_fitness": 0.8184538453768972,
    "target_cdf": 0.9877755264344382,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.548961639404297,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7677538905949514,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.548961639404297,
        "normalized": 0.9877755264344382,
        "contribution": 0.4938877632172191
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.48657751083374,
        "normalized": 0.9534909337644679,
        "contribution": 0.1430236400646702
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7480251789093018,
        "normalized": 0.877800656022249,
        "contribution": 0.0877800656022249
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "da2e5cc444ed3d66978985fe0306f2a6cce1fc3954134e05829711aa9124388c",
    "id": "334f551eec499513a0c898357b60",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.381826400756836,
    "wer": 0.0,
    "identity": 0.6422304511070251,
    "voiceclap": -0.09432007372379303,
    "fitness": 0.76045504934754,
    "legacy_fitness": 0.76045504934754,
    "target_cdf": 0.8450960105094518,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2665977478027344,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6547807660890479,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2665977478027344,
        "normalized": 0.8450960105094518,
        "contribution": 0.4225480052547259
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2857985496520996,
        "normalized": 0.3231226457917462,
        "contribution": 0.08078066144793655
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.7741496562957764,
        "normalized": 0.5697670759266259,
        "contribution": 0.08546506138899389
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.900927782058716,
        "normalized": 0.6598703799739153,
        "contribution": 0.06598703799739153
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "49d838b7daeffb2001a03259c0612644a1a94b7329d8d7c51ff65cd7127869d3",
    "id": "33709f3d2b34c7fb464140615310",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.6663577556610107,
    "wer": 0.0,
    "identity": 0.5553252696990967,
    "voiceclap": 0.08091993629932404,
    "fitness": 0.7835249064307138,
    "legacy_fitness": 0.7835249064307138,
    "target_cdf": 0.9445039705317666,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8671984672546387,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7564945176437967,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8671984672546387,
        "normalized": 0.9445039705317666,
        "contribution": 0.4722519852658833
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1144819259643555,
        "normalized": 0.25583161582232095,
        "contribution": 0.06395790395558024
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8096323013305664,
        "normalized": 0.8885304573103883,
        "contribution": 0.08885304573103883
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "595cb740376b5ed8b457fd45f39a571cfb74a0dd06a2c294ed4e3a878b6e7bb5",
    "id": "33aaa1f6361244d30b35aff77344",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 3.8044304847717285,
    "wer": 0.0,
    "identity": 0.6330751180648804,
    "voiceclap": 0.0073515549302101135,
    "fitness": 0.7775296890008541,
    "legacy_fitness": 0.7775296890008541,
    "target_cdf": 0.9178304042951448,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6561765670776367,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7030927483667123,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6561765670776367,
        "normalized": 0.9178304042951448,
        "contribution": 0.4589152021475724
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.951646327972412,
        "normalized": 0.19940455387819955,
        "contribution": 0.04985113846954989
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5792131423950195,
        "normalized": 0.8113311992293513,
        "contribution": 0.12169967988440268
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.114352226257324,
        "normalized": 0.7262672786518737,
        "contribution": 0.07262672786518737
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f12956da3e3dce73f8b8f8ca48172b8e04d88c1e30d5e3552864820815a1555c",
    "id": "33b29e17d41cbf2f45952dd04140",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.68307638168335,
    "wer": 0.0,
    "identity": 0.5545382499694824,
    "voiceclap": -0.1004551500082016,
    "fitness": 0.799337077032132,
    "legacy_fitness": 0.799337077032132,
    "target_cdf": 0.895770231665379,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5182008743286133,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6705945241812521,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5182008743286133,
        "normalized": 0.895770231665379,
        "contribution": 0.4478851158326895
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0296716690063477,
        "normalized": 0.2254527264287831,
        "contribution": 0.05636318160719578
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.000990867614746,
        "normalized": 0.6461697861380791,
        "contribution": 0.09692546792071186
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.008740186691284,
        "normalized": 0.6942075882065499,
        "contribution": 0.069420758820655
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e8a3f62919b4b8cf886a586d1a20f764ce33b424b765caad09ef2eb889ec41ba",
    "id": "340107472d4316ca647d547cc8c7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.425758361816406,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.027339890599250793,
    "fitness": 0.8378769000173742,
    "legacy_fitness": 0.8378769000173742,
    "target_cdf": 0.9887331723771762,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5814266204833984,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7720596895466623,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5814266204833984,
        "normalized": 0.9887331723771762,
        "contribution": 0.4943665861885881
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.066988229751587,
        "normalized": 0.2385614708758641,
        "contribution": 0.059640367718966025
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.086269855499268,
        "normalized": 0.9079316126330063,
        "contribution": 0.13618974189495092
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4619922637939453,
        "normalized": 0.8186299374415724,
        "contribution": 0.08186299374415724
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "87a553c74328c1ed1458cbaa41ca652bed02d03e7e16432b88fd4ab82338fed7",
    "id": "3457946d28a8ce52abafeee7939b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": 0.7233633995056152,
    "voiceclap": -0.06987233459949493,
    "fitness": 0.7691208345361431,
    "legacy_fitness": 0.7691208345361431,
    "target_cdf": 0.8793763603106894,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.428922176361084,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6978985387223742,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.428922176361084,
        "normalized": 0.8793763603106894,
        "contribution": 0.4396881801553447
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.094127655029297,
        "normalized": 0.2483512185262678,
        "contribution": 0.06208780463156695
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4813601970672607,
        "normalized": 0.7872381937406205,
        "contribution": 0.11808572906109308
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3079745769500732,
        "normalized": 0.780368248743694,
        "contribution": 0.07803682487436941
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b010b3301f946f4328278af577e168dabcd807970bf24bce77fdef2d929ed034",
    "id": "3466b14370686a1a0fa47e83b0c2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.086852550506592,
    "wer": 0.0,
    "identity": 0.6021595001220703,
    "voiceclap": -0.06654146313667297,
    "fitness": 0.7694745523400981,
    "legacy_fitness": 0.7694745523400981,
    "target_cdf": 0.8793763603106894,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.428922176361084,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.666401501081387,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.428922176361084,
        "normalized": 0.8793763603106894,
        "contribution": 0.4396881801553447
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8736209869384766,
        "normalized": 0.17525676081671665,
        "contribution": 0.04381419020417916
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.352372169494629,
        "normalized": 0.7528841549820324,
        "contribution": 0.11293262324730485
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0263421535491943,
        "normalized": 0.6996650747455826,
        "contribution": 0.06996650747455826
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1b5d22af81763efb5549c08b430bd617c524ebc4f0b666333e03cdc579052381",
    "id": "35968750175c532cfd728c99ad16",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.197352409362793,
    "genuineness": 4.174716949462891,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.002919662743806839,
    "fitness": 0.8278702809123931,
    "legacy_fitness": 0.8278702809123931,
    "target_cdf": 0.9979798622347391,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1982598304748535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7642838490965356,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1982598304748535,
        "normalized": 0.9979798622347391,
        "contribution": 0.49898993111736956
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7277473211288452,
        "normalized": 0.1353259560349948,
        "contribution": 0.0338314890087487
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.477681636810303,
        "normalized": 0.9527255009885056,
        "contribution": 0.14290882514827583
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7920303344726562,
        "normalized": 0.8855360382214144,
        "contribution": 0.08855360382214145
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "be91fdac48471cce3ce58b168ae26a83f36e54ee1afb7bc15fcf4885ea138d53",
    "id": "361eb95ccad12dce8c00b7e87536",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 3,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03112182207405567,
    "fitness": 0.7959357246364169,
    "legacy_fitness": 0.7959357246364169,
    "target_cdf": 0.9453737983316208,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.875314712524414,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7737530403256603,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.875314712524414,
        "normalized": 0.9453737983316208,
        "contribution": 0.4726868991658104
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2942795753479004,
        "normalized": 0.3266368847639896,
        "contribution": 0.0816592211909974
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.935042381286621,
        "normalized": 0.8840175306994371,
        "contribution": 0.13260262960491556
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.695219039916992,
        "normalized": 0.8680429036393686,
        "contribution": 0.08680429036393687
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "441c62b03b6bed7ab188b52437c63a0cb8b0842d24ce3c83c0a52315b8437834",
    "id": "361f3e26a650612a690cb4ae69cd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.02204098366200924,
    "fitness": 0.8196326263079908,
    "legacy_fitness": 0.8196326263079908,
    "target_cdf": 0.9985792095124457,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.311887264251709,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8444755961668495,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.311887264251709,
        "normalized": 0.9985792095124457,
        "contribution": 0.49928960475622286
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4503304958343506,
        "normalized": 0.3937604400617085,
        "contribution": 0.09844011001542713
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.367254734039307,
        "normalized": 0.9929189456061928,
        "contribution": 0.1489378418409289
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.707335472106934,
        "normalized": 0.9780803955427062,
        "contribution": 0.09780803955427063
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "632e429a6e04d8ce8991f1f9afec62d621b6c52ec6bc3127fa9c706ddd5a1574",
    "id": "36a86ef06681c049c28e7d65ab0d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.0146708488464355,
    "genuineness": 4.18726921081543,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.14990058541297913,
    "fitness": 0.8116652646769378,
    "legacy_fitness": 0.8116652646769378,
    "target_cdf": 0.9736578860102679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2243127822875977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7445329219632866,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2243127822875977,
        "normalized": 0.9736578860102679,
        "contribution": 0.48682894300513396
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.979520797729492,
        "normalized": 0.8914683718087433,
        "contribution": 0.1337202557713115
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4223875999450684,
        "normalized": 0.8092130147600393,
        "contribution": 0.08092130147600393
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "359e1f7ca09ecd606cdbc870509c4f542d6e3abbb0d54c983e1c5449d35e4e42",
    "id": "36ac6b4b646283d1ab9b78aa7733",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03916459158062935,
    "fitness": 0.8245825046011546,
    "legacy_fitness": 0.8245825046011546,
    "target_cdf": 0.9979798622347391,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1982598304748535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7610195942542667,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1982598304748535,
        "normalized": 0.9979798622347391,
        "contribution": 0.49898993111736956
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7277473211288452,
        "normalized": 0.1353259560349948,
        "contribution": 0.0338314890087487
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.495473384857178,
        "normalized": 0.9542463876094935,
        "contribution": 0.14313695814142402
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6072089672088623,
        "normalized": 0.8506121598672445,
        "contribution": 0.08506121598672445
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2aa79f3b7df7d1e6364ed719d5a5deefbf16eef79299c775bf59f33c5a117b74",
    "id": "3710f6b41280d51a11dabb303c52",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.026847369968891144,
    "fitness": 0.8171667054362154,
    "legacy_fitness": 0.8171667054362154,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7267414766287821,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.361367106437683,
        "normalized": 0.06382463962327856,
        "contribution": 0.01595615990581964
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.255288600921631,
        "normalized": 0.9301035297844544,
        "contribution": 0.13951552946766815
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2199645042419434,
        "normalized": 0.7565702736046067,
        "contribution": 0.07565702736046068
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a888ae893c8cb7c591eff18ce1486111ac25fe5f829dbe5380726eaedde5ae44",
    "id": "37f334b5d705dd6149e23f037f45",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.6412534713745117,
    "wer": 0.0,
    "identity": 0.7431926727294922,
    "voiceclap": 0.053909145295619965,
    "fitness": 0.7966587184974843,
    "legacy_fitness": 0.7966587184974843,
    "target_cdf": 0.9736578860102679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2243127822875977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7252640234840144,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2243127822875977,
        "normalized": 0.9736578860102679,
        "contribution": 0.48682894300513396
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6056206226348877,
        "normalized": 0.10708370011583324,
        "contribution": 0.02677092502895831
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.241966962814331,
        "normalized": 0.762647195792524,
        "contribution": 0.0762647195792524
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0e2ed82918d959d4d990f3212e9928799e3aea488142b78f84a4c613161f86e6",
    "id": "38c31d28d368861e206680db652b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.22160956263542175,
    "fitness": 0.8241589063617241,
    "legacy_fitness": 0.8241589063617241,
    "target_cdf": 0.9599408456112248,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0295231342315674,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7214031910754723,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0295231342315674,
        "normalized": 0.9599408456112248,
        "contribution": 0.4799704228056124
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9652159214019775,
        "normalized": 0.20379983538139917,
        "contribution": 0.050949958845349794
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5614216327667236,
        "normalized": 0.8070803373590341,
        "contribution": 0.12106205060385511
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.008740186691284,
        "normalized": 0.6942075882065499,
        "contribution": 0.069420758820655
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7f513ea753f35684117c9fd8fa6b7bb6505c50c8813ea46499df352d4bca0868",
    "id": "390efd3da3a091b2192accc27164",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": 0.7135661244392395,
    "voiceclap": -0.06328413635492325,
    "fitness": 0.712720399279733,
    "legacy_fitness": 0.712720399279733,
    "target_cdf": 0.7391168299900283,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.8770188093185425,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5715476877425103,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.8770188093185425,
        "normalized": 0.7391168299900283,
        "contribution": 0.36955841499501413
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.104304790496826,
        "normalized": 0.2520767788636856,
        "contribution": 0.0630191947159214
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.653501510620117,
        "normalized": 0.5278332519667204,
        "contribution": 0.07917498779500805
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.716106414794922,
        "normalized": 0.5979509023656668,
        "contribution": 0.059795090236566685
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c77154e4c5da4fe1561f0a5f0726f3d52d617b14ce8d25275f0f8e388509da64",
    "id": "39655e0400bedb5cccc0dbf5eee8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04832199960947037,
    "fitness": 0.8229684744516804,
    "legacy_fitness": 0.8229684744516804,
    "target_cdf": 0.9832067001342365,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.419102191925049,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7289662526657363,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.419102191925049,
        "normalized": 0.9832067001342365,
        "contribution": 0.49160335006711825
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6124054193496704,
        "normalized": 0.10853160036418047,
        "contribution": 0.027132900091045117
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1495563983917236,
        "normalized": 0.7365719667634352,
        "contribution": 0.07365719667634352
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "560889c9e8facbf347809b3081cab5307d38fa4b651ffc61187d2f17c1285804",
    "id": "39d16dfd8a8147751857a2b05b9f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": 0.5732236504554749,
    "voiceclap": -0.0367620624601841,
    "fitness": 0.8284851862198228,
    "legacy_fitness": 0.8284851862198228,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7904855599792528,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2400012016296387,
        "normalized": 0.3044239560639862,
        "contribution": 0.07610598901599655
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.006208419799805,
        "normalized": 0.8957702550325092,
        "contribution": 0.13436553825487638
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7480251789093018,
        "normalized": 0.877800656022249,
        "contribution": 0.0877800656022249
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "dd5ee3b60a2a9b828c9ecdbb3dbf00498da339eccaea3e06023491e7c7918edd",
    "id": "39e18e06d53821379a73109c67fd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.05093712732195854,
    "fitness": 0.8115695129452318,
    "legacy_fitness": 0.8115695129452318,
    "target_cdf": 0.9495623609658421,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.915895938873291,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7416151218821012,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.915895938873291,
        "normalized": 0.9495623609658421,
        "contribution": 0.47478118048292106
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0907351970672607,
        "normalized": 0.24711589066920464,
        "contribution": 0.06177897266730116
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.637035369873047,
        "normalized": 0.8247432450396308,
        "contribution": 0.12371148675594462
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4399898052215576,
        "normalized": 0.8134348197593442,
        "contribution": 0.08134348197593443
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "678c50a4bd24d408462b02b48aaeef022fb25f0bbe4aba420074d63ede464643",
    "id": "3a16eed66c144d3919e912a48555",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": 0.4847373366355896,
    "voiceclap": 0.017856406047940254,
    "fitness": 0.8219223278011654,
    "legacy_fitness": 0.8219223278011654,
    "target_cdf": 0.996049248624995,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.971005439758301,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8308492929005445,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.971005439758301,
        "normalized": 0.996049248624995,
        "contribution": 0.4980246243124975
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3892672061920166,
        "normalized": 0.36698712459557925,
        "contribution": 0.09174678114889481
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.620013236999512,
        "normalized": 0.9638222528181529,
        "contribution": 0.14457333792272292
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.478508949279785,
        "normalized": 0.9650454951642935,
        "contribution": 0.09650454951642935
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "953a073445ca01ae998c38dfbe34e750d38fd5f67f51c9758abce2c7cf43ea33",
    "id": "3a2588883f92f8c5983347ce03ea",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.008801953867077827,
    "fitness": 0.8232045577704693,
    "legacy_fitness": 0.8232045577704693,
    "target_cdf": 0.9862070255954634,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5002641677856445,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7798700913671847,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5002641677856445,
        "normalized": 0.9862070255954634,
        "contribution": 0.4931035127977317
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0873427391052246,
        "normalized": 0.2458838571357379,
        "contribution": 0.061470964283934476
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7392241954803467,
        "normalized": 0.8762105466288791,
        "contribution": 0.08762105466288791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "178e36c665dcf49ab015a91eabfe7a41840ee176af418d44c5fac912a240d9e1",
    "id": "3a341a5df18a6e6092b35bb3d276",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": 0.5969081521034241,
    "voiceclap": -0.06704128533601761,
    "fitness": 0.8367156619534231,
    "legacy_fitness": 0.8367156619534231,
    "target_cdf": 0.9929189433068706,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.759983539581299,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7206867624647901,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.759983539581299,
        "normalized": 0.9929189433068706,
        "contribution": 0.4964594716534353
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.766023635864258,
        "normalized": 0.8524214149132157,
        "contribution": 0.12786321223698235
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.821718692779541,
        "normalized": 0.6337481353490892,
        "contribution": 0.06337481353490892
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "de8dc9aec45ede16b41dbe6b34a09629cc5a6af46e84908dd2c6c80a8d927fc6",
    "id": "3a5fc5c82aef17ba4fe4cee11322",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.005938416346907616,
    "fitness": 0.8187482782809449,
    "legacy_fitness": 0.8187482782809449,
    "target_cdf": 0.9989025172806805,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.393049240112305,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.775947993614504,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.393049240112305,
        "normalized": 0.9989025172806805,
        "contribution": 0.49945125864034023
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8498740196228027,
        "normalized": 0.16829178263767364,
        "contribution": 0.04207294565941841
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.691179275512695,
        "normalized": 0.9685129990310137,
        "contribution": 0.14527694985465206
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8272342681884766,
        "normalized": 0.891468394600932,
        "contribution": 0.0891468394600932
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d79c6ebfffdc655426b9f71ddf511cbb877a73c61abfacab1262738d5aa9909c",
    "id": "3a7ebabf9aa94038b9cd137056fc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.469690799713135,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10574140399694443,
    "fitness": 0.8289572407017343,
    "legacy_fitness": 0.8289572407017343,
    "target_cdf": 0.9656442819456641,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.102569103240967,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6997254126663299,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.102569103240967,
        "normalized": 0.9656442819456641,
        "contribution": 0.48282214097283205
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5658695697784424,
        "normalized": 0.8081485136597149,
        "contribution": 0.12122227704895722
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8987274169921875,
        "normalized": 0.6591543062832207,
        "contribution": 0.06591543062832207
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "dd985bc3ef6afd05393d861ebdcec5261ab6e7ef2fb03b4a543642b30ca17eeb",
    "id": "3ac93088c28cc156693e7f5b70c3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.562715530395508,
    "genuineness": 4.375550270080566,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03217737376689911,
    "fitness": 0.8392059690350593,
    "legacy_fitness": 0.8392059690350593,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7089893246752239,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4292153120040894,
        "normalized": 0.0741850635010084,
        "contribution": 0.0185462658752521
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.068478107452393,
        "normalized": 0.9053242349151492,
        "contribution": 0.13579863523727237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7755134105682373,
        "normalized": 0.6182093737421804,
        "contribution": 0.061820937374218045
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0bda1a13194ca92b07e6d66de811d1115db5d39641c93c8106433416f2ba0b49",
    "id": "3c027ebbe3cbd66be244c961b1e8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.1020402908325195,
    "genuineness": 3.5847692489624023,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.18941086530685425,
    "fitness": 0.8018441523414477,
    "legacy_fitness": 0.8018441523414477,
    "target_cdf": 0.9998555193944024,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.977417945861816,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8618876353577568,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.977417945861816,
        "normalized": 0.9998555193944024,
        "contribution": 0.4999277596972012
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.6169826984405518,
        "normalized": 0.4690585513651788,
        "contribution": 0.1172646378412947
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.2427144050598145,
        "normalized": 0.9904545174589181,
        "contribution": 0.1485681776188377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.425703048706055,
        "normalized": 0.9612706020042315,
        "contribution": 0.09612706020042316
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "741ebdb580181762d06a5794b17d546ba566a5c64ff538664b7df04eb981ca07",
    "id": "3c5feb1307a2b3efb47e06647c89",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.086852550506592,
    "wer": 0.0,
    "identity": 0.6537616848945618,
    "voiceclap": -0.0005235821008682251,
    "fitness": 0.7979941317110415,
    "legacy_fitness": 0.7979941317110415,
    "target_cdf": 0.9399897234974615,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8266172409057617,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7161546769871792,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8266172409057617,
        "normalized": 0.9399897234974615,
        "contribution": 0.46999486174873073
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9720008373260498,
        "normalized": 0.206019001985337,
        "contribution": 0.05150475049633425
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.570317506790161,
        "normalized": 0.8092130613886472,
        "contribution": 0.12138195920829709
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.136354923248291,
        "normalized": 0.7327310553381711,
        "contribution": 0.0732731055338171
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "31e05a723ec3adcd1fa5e95ce671067bbdd0f4fa18135fa865f77bbbd7149358",
    "id": "3c64f3279fa823407a8f872fac48",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.08728805929422379,
    "fitness": 0.8296611149220956,
    "legacy_fitness": 0.8296611149220956,
    "target_cdf": 0.9736578860102679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2243127822875977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7257292773986608,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2243127822875977,
        "normalized": 0.9736578860102679,
        "contribution": 0.48682894300513396
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7820258140563965,
        "normalized": 0.14938781636553056,
        "contribution": 0.03734695409138264
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8016064167022705,
        "normalized": 0.8595110162463788,
        "contribution": 0.1289266524369568
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.114352226257324,
        "normalized": 0.7262672786518737,
        "contribution": 0.07262672786518737
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "12c15f67d26069083bc371da9faf39621c1b1d6aee31ac7372cc3d7cea4b5597",
    "id": "3ca0f66c3c1c821c8984b8dd8949",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": 0.24410046637058258,
    "voiceclap": 0.05099421739578247,
    "fitness": 0.827488577423964,
    "legacy_fitness": 0.827488577423964,
    "target_cdf": 0.9993562461107455,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.555373668670654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8182251693417846,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.555373668670654,
        "normalized": 0.9993562461107455,
        "contribution": 0.49967812305537274
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2298238277435303,
        "normalized": 0.3003348835264067,
        "contribution": 0.07508372088160167
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.224923133850098,
        "normalized": 0.9900472697166317,
        "contribution": 0.14850709045749474
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.284886837005615,
        "normalized": 0.9495623494731538,
        "contribution": 0.09495623494731538
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cba9bbea7ffe929ee02dde99f7f93dfa27d5f1205b2b7f7b977b5bf8f13d1e3c",
    "id": "3ce9e99413f35c3b1f1962fd58dd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04978766292333603,
    "fitness": 0.8243765292290899,
    "legacy_fitness": 0.8243765292290899,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8068585735690035,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3943557739257812,
        "normalized": 0.36919567975921314,
        "contribution": 0.09229891993980328
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.862438201904297,
        "normalized": 0.897176290990003,
        "contribution": 0.0897176290990003
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "594c26ed2ca150d7277610b2788dd4b7b8636a12e020d09bc329b0b6ab9742f2",
    "id": "3d4425e24ee14a5246f77b3bab2e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07553695887327194,
    "fitness": 0.7737119673084671,
    "legacy_fitness": 0.7737119673084671,
    "target_cdf": 0.8929163174612232,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5019683837890625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7753844868680237,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5019683837890625,
        "normalized": 0.8929163174612232,
        "contribution": 0.4464581587306116
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.641153573989868,
        "normalized": 0.4801390317750176,
        "contribution": 0.1200347579437544
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6459312438964844,
        "normalized": 0.826751667074274,
        "contribution": 0.1240127500611411
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5984079837799072,
        "normalized": 0.8487882013251657,
        "contribution": 0.08487882013251657
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0b0fa92a5fe778680df75b1a0a75a67909a5f07a9dfb23c017983c99c60ab774",
    "id": "3d54aca08082a71b89fe7c108c23",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.432034492492676,
    "wer": 0.0,
    "identity": 0.49082815647125244,
    "voiceclap": -0.12900467216968536,
    "fitness": 0.7857034442726525,
    "legacy_fitness": 0.7857034442726525,
    "target_cdf": 0.8870403206157192,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.469503402709961,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7320127829838678,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.469503402709961,
        "normalized": 0.8870403206157192,
        "contribution": 0.4435201603078596
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4435458183288574,
        "normalized": 0.3907581380976326,
        "contribution": 0.09768953452440815
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2589669227600098,
        "normalized": 0.7262672960496188,
        "contribution": 0.10894009440744282
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4619922637939453,
        "normalized": 0.8186299374415724,
        "contribution": 0.08186299374415724
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "14c8d0313fc44a5e2774af6d251dd099f022b91e52b9c215350fe5964e28e5e6",
    "id": "3d7cde80d71f09ea713a3382fe3f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.08978249132633209,
    "fitness": 0.8361006992781215,
    "legacy_fitness": 0.8361006992781215,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7157710839616643,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.601452350616455,
        "normalized": 0.8165628197935635,
        "contribution": 0.12248442296903451
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.845921516418457,
        "normalized": 0.6418026300201901,
        "contribution": 0.06418026300201901
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4c06d0baafe6060a6de6e37cd329de6c1f142eeb5de8bb2772bbf8d2899f083a",
    "id": "3ddcf074f77f62347b7a94f0a42a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.20609712600708,
    "wer": 0.0,
    "identity": 0.5788609981536865,
    "voiceclap": -0.14657163619995117,
    "fitness": 0.7896986609437815,
    "legacy_fitness": 0.7896986609437815,
    "target_cdf": 0.9166388046222431,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6480603218078613,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6549877718736951,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6480603218078613,
        "normalized": 0.9166388046222431,
        "contribution": 0.45831940231112156
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8091652393341064,
        "normalized": 0.1567722644593298,
        "contribution": 0.03919306611483245
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.9676318168640137,
        "normalized": 0.6352170794829098,
        "contribution": 0.09528256192243648
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7865145206451416,
        "normalized": 0.6219274152530461,
        "contribution": 0.06219274152530461
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c6e8f0eb6ac25d67dd6bed7c3cb2e2ec99a297a47038688d7e278f48f1033fa2",
    "id": "3de02683255ad6a2a2787245eaff",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.515059471130371,
    "genuineness": 4.049196243286133,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.08936218917369843,
    "fitness": 0.8244164831036629,
    "legacy_fitness": 0.8244164831036629,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7373374044932948,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.121852874755859,
        "normalized": 0.9129861397194918,
        "contribution": 0.13694792095792377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3431787490844727,
        "normalized": 0.7895000148839948,
        "contribution": 0.07895000148839948
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "33837ff3a08722b492a6c69e433f3c1eec614d98e0a064a4f1ff1b714664498d",
    "id": "3e377ea9df8c0487e9a581a62cbd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.06220494583249092,
    "fitness": 0.8188970156147555,
    "legacy_fitness": 0.8188970156147555,
    "target_cdf": 0.9982603445916324,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.246957302093506,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8605094849030526,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.246957302093506,
        "normalized": 0.9982603445916324,
        "contribution": 0.4991301722958162
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.596628189086914,
        "normalized": 0.45974578206657163,
        "contribution": 0.11493644551664291
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.385046005249023,
        "normalized": 0.9932207029626189,
        "contribution": 0.14898310544439283
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.636927604675293,
        "normalized": 0.974597616462006,
        "contribution": 0.09745976164620061
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "aa64e709dfd2b26923c031322bafb110cbfe4aa7cfcda03003105022744fb41b",
    "id": "3e8bff7323e2e376cd2a4dcf561e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 4.011539936065674,
    "wer": 0.0,
    "identity": 0.6413116455078125,
    "voiceclap": 0.07221878319978714,
    "fitness": 0.8099642139220666,
    "legacy_fitness": 0.8099642139220666,
    "target_cdf": 0.9662346008508267,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.110685348510742,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7178380470238014,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.110685348510742,
        "normalized": 0.9662346008508267,
        "contribution": 0.48311730042541334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.748231887817383,
        "normalized": 0.8487881729410258,
        "contribution": 0.12731822594115386
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1319541931152344,
        "normalized": 0.7314444306568307,
        "contribution": 0.07314444306568307
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "54fdb00ece9d625f5724032eaeda1600d4eee3d3937aa79a8e0c8f49fbc666ba",
    "id": "3e995f6a03e9abd61a2d59991892",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.860914945602417,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.035781584680080414,
    "fitness": 0.8045844146818113,
    "legacy_fitness": 0.8045844146818113,
    "target_cdf": 0.9656442819456641,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.102569103240967,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.755926012671025,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.102569103240967,
        "normalized": 0.9656442819456641,
        "contribution": 0.48282214097283205
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9787856340408325,
        "normalized": 0.2082523969786677,
        "contribution": 0.05206309924466693
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.068478107452393,
        "normalized": 0.9053242349151492,
        "contribution": 0.13579863523727237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6160099506378174,
        "normalized": 0.852421372162536,
        "contribution": 0.08524213721625361
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8be29616e66fad4234b2775db778b6fe77222172e9cd3545cba8fe750f46ebc3",
    "id": "3f13ce3a8cfdbd2f0c5cf5c4ed30",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": 0.11727505922317505,
    "voiceclap": 0.03708869218826294,
    "fitness": 0.7835663539834697,
    "legacy_fitness": 0.7835663539834697,
    "target_cdf": 0.8793763603106894,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.428922176361084,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7225613990670846,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.428922176361084,
        "normalized": 0.8793763603106894,
        "contribution": 0.4396881801553447
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.503761053085327,
        "normalized": 0.417611474538243,
        "contribution": 0.10440286863456075
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2011446952819824,
        "normalized": 0.7091081777136116,
        "contribution": 0.10636622665704173
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.096750259399414,
        "normalized": 0.721041236201374,
        "contribution": 0.07210412362013739
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "dff7fbfc4a343a3cf6e9acbf9f4242c06b430fce0afd81670faeec7ac8eebea5",
    "id": "3f275637f37552b1ec2cf2662ca2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04617062211036682,
    "fitness": 0.7835374196059202,
    "legacy_fitness": 0.7835374196059202,
    "target_cdf": 0.8824847618116329,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4451546669006348,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6429592100862076,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4451546669006348,
        "normalized": 0.8824847618116329,
        "contribution": 0.44124238090581647
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.178905487060547,
        "normalized": 0.7023772573695775,
        "contribution": 0.10535658860543662
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7711129188537598,
        "normalized": 0.6167190605492896,
        "contribution": 0.06167190605492896
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2327f6a53402f2a7fe5ca221739128ab992b7203f6ee50e035742a21f58e3018",
    "id": "3f2b020df8f85c9cd39e6e87233d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.260893821716309,
    "genuineness": 4.293961524963379,
    "wer": 0.03571428571428571,
    "identity": 0.5837700366973877,
    "voiceclap": -0.004437088966369629,
    "fitness": 0.7896011947069669,
    "legacy_fitness": 0.7896011947069669,
    "target_cdf": 0.9668165486335377,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1188015937805176,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7530881226439228,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1188015937805176,
        "normalized": 0.9668165486335377,
        "contribution": 0.48340827431676886
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1416213512420654,
        "normalized": 0.26598556001729745,
        "contribution": 0.06649639000432436
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5169429779052734,
        "normalized": 0.7962001195838251,
        "contribution": 0.11943001793757375
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5456018447875977,
        "normalized": 0.8375344038525582,
        "contribution": 0.08375344038525583
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3aecdab5b2eae425283df9cdcee3900d68d0d82499f8c97ad49c1485eae5e397",
    "id": "3f308afa110fe7f6b19d31e361bc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.18726921081543,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07066242396831512,
    "fitness": 0.8040306955410791,
    "legacy_fitness": 0.8040306955410791,
    "target_cdf": 0.9409148593413338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.834733486175537,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6969501302835557,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.834733486175537,
        "normalized": 0.9409148593413338,
        "contribution": 0.4704574296706669
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.768456220626831,
        "normalized": 0.1457841285203917,
        "contribution": 0.03644603213009792
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.632587432861328,
        "normalized": 0.8237335195645583,
        "contribution": 0.12356002793468374
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9163296222686768,
        "normalized": 0.6648664054810712,
        "contribution": 0.06648664054810713
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "57fb3eff51e85d08ecb26992b9c7643203d33257b44734978048840334ac7ec0",
    "id": "3f70c7b46bedd5191311d08e4316",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.086852550506592,
    "wer": 0.0,
    "identity": 0.7603334784507751,
    "voiceclap": -0.11955438554286957,
    "fitness": 0.8179358733100408,
    "legacy_fitness": 0.8179358733100408,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7583100499803181,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9448615312576294,
        "normalized": 0.19722848067005766,
        "contribution": 0.049307120167514415
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8016064167022705,
        "normalized": 0.8595110162463788,
        "contribution": 0.1289266524369568
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8272342681884766,
        "normalized": 0.891468394600932,
        "contribution": 0.0891468394600932
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1157365dbb8f97ca40ee885bf9f54332074da9d503cf27987a9aed6af5266a97",
    "id": "3f78e1aaaa3271eeae951355f5bf",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.973883628845215,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": -0.20345604419708252,
    "fitness": 0.7328799716058461,
    "legacy_fitness": 0.7328799716058461,
    "target_cdf": 0.8629676751891268,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.347759962081909,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6770373347812746,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.347759962081909,
        "normalized": 0.8629676751891268,
        "contribution": 0.4314838375945634
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.033064126968384,
        "normalized": 0.2266273368759334,
        "contribution": 0.05665683421898335
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3568201065063477,
        "normalized": 0.7541161908392326,
        "contribution": 0.11311742862588489
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.224364995956421,
        "normalized": 0.7577923434184293,
        "contribution": 0.07577923434184293
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d92de1ab3542475b0e02fba5361c4ebc84ffe2d8ae3aa05223714a3701a3f7a2",
    "id": "3f8b45bd11c7dbc6b5678c9e884b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 5.974957466125488,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07417355477809906,
    "fitness": 0.7844371383029933,
    "legacy_fitness": 0.7844371383029933,
    "target_cdf": 0.9154342620847058,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.639944076538086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7510804711865967,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.639944076538086,
        "normalized": 0.9154342620847058,
        "contribution": 0.4577171310423529
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4935836791992188,
        "normalized": 0.41304203347797125,
        "contribution": 0.10326050836949281
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4012985229492188,
        "normalized": 0.7662529752285923,
        "contribution": 0.11493794628428883
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.202362537384033,
        "normalized": 0.7516488549046221,
        "contribution": 0.07516488549046221
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "218cd98d03d0b0c5146550fcc55049e49253ad55de5b5292c3f4658f4ecee769",
    "id": "3ff69ff213ae1d4ff4051148e20b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.948779582977295,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0772237479686737,
    "fitness": 0.823455630256142,
    "legacy_fitness": 0.823455630256142,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7879286191915627,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9652159214019775,
        "normalized": 0.20379983538139917,
        "contribution": 0.050949958845349794
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.993634223937988,
        "normalized": 0.9832067001034072,
        "contribution": 0.14748100501551106
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9944534301757812,
        "normalized": 0.9166388033350146,
        "contribution": 0.09166388033350147
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "88393de77970e85d86223578dd39f62c9c3e8979b3cf88c93c09901c70666146",
    "id": "4035cf4c42ee66fcef724567b16c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.6203155517578125,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.005671467632055283,
    "fitness": 0.8419522592045576,
    "legacy_fitness": 0.8419522592045576,
    "target_cdf": 0.9838478622074346,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4353346824645996,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7417889093580197,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4353346824645996,
        "normalized": 0.9838478622074346,
        "contribution": 0.4919239311037173
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.948253870010376,
        "normalized": 0.19831469683683772,
        "contribution": 0.04957867420920943
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.854980707168579,
        "normalized": 0.8697054717721656,
        "contribution": 0.13045582076582485
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.021941661834717,
        "normalized": 0.6983048327926806,
        "contribution": 0.06983048327926807
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4f4706a8ecd10c595cdd1b200dd1f5dfcac6b9365d403ad657f8d492da0bd0d2",
    "id": "4045f7516f2d802e50644cb3e0d6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.078212261199951,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.16780436038970947,
    "fitness": 0.7936265045373213,
    "legacy_fitness": 0.7936265045373213,
    "target_cdf": 0.9213282215532803,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.680525302886963,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6631499101173343,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.680525302886963,
        "normalized": 0.9213282215532803,
        "contribution": 0.46066411077664016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8837981224060059,
        "normalized": 0.17829687836447006,
        "contribution": 0.044574219591117514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.07438063621521,
        "normalized": 0.6698334010415226,
        "contribution": 0.10047501015622838
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6478986740112305,
        "normalized": 0.574365695933482,
        "contribution": 0.0574365695933482
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "39c625ee029ecddec6ad0aa573ae3368f13888930fc2859659e548ce898be0f3",
    "id": "4115b28c1c3e340d9e668d5ff642",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.149612903594971,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.08834057301282883,
    "fitness": 0.8314265514546689,
    "legacy_fitness": 0.8314265514546689,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7757269975562133,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9007601737976074,
        "normalized": 0.18343713514349452,
        "contribution": 0.04585928378587363
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.620013236999512,
        "normalized": 0.9638222528181529,
        "contribution": 0.14457333792272292
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7304232120513916,
        "normalized": 0.8746060085041646,
        "contribution": 0.08746060085041646
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "78e5c3c04854a95850e4259df911e642ae33742e6a38571a5597d5464ffb5f40",
    "id": "41810833cd3494e92696e36dc66d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.268857479095459,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09792295098304749,
    "fitness": 0.7993985312586965,
    "legacy_fitness": 0.7993985312586965,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7041445025632422,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8905830383300781,
        "normalized": 0.18034202015435835,
        "contribution": 0.04508550503858959
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.539182424545288,
        "normalized": 0.8016852878192227,
        "contribution": 0.1202527931728834
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.290372610092163,
        "normalized": 0.7757183970251897,
        "contribution": 0.07757183970251898
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "112ae4605f7e16c609ae600ac331f95eadfab0d57a0a9e6a7f391528832ec9bc",
    "id": "4184f1b84e03a9fc74d6cc72306a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": 0.6347166299819946,
    "voiceclap": -0.011806624010205269,
    "fitness": 0.8119113954991906,
    "legacy_fitness": 0.8119113954991906,
    "target_cdf": 0.9811510787587977,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3704047203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.730222784847254,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3704047203063965,
        "normalized": 0.9811510787587977,
        "contribution": 0.49057553937939885
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6259750127792358,
        "normalized": 0.11146953771947371,
        "contribution": 0.027867384429868428
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.952833652496338,
        "normalized": 0.8870402984992682,
        "contribution": 0.13305604477489022
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3343775272369385,
        "normalized": 0.787238162630965,
        "contribution": 0.0787238162630965
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7ca1fc30478c6a248faf423cf3323e9cd7470f6afa810a65f1ff7001d9490396",
    "id": "41d90dc5c8cf43b7ad97d621a325",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.036643981933594,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.11901092529296875,
    "fitness": 0.8283683992859484,
    "legacy_fitness": 0.8283683992859484,
    "target_cdf": 0.9981710548076552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.230724811553955,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7474491635156061,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.230724811553955,
        "normalized": 0.9981710548076552,
        "contribution": 0.4990855274038276
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5174179077148438,
        "normalized": 0.0895146120757892,
        "contribution": 0.0223786530189473
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.691179275512695,
        "normalized": 0.9685129990310137,
        "contribution": 0.14527694985465206
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4135866165161133,
        "normalized": 0.807080332381791,
        "contribution": 0.0807080332381791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a76724146b62f8493c19288cd0df31a083464db8b27a8f5c576bbaca0bf035ff",
    "id": "41ffcc88e3e9ad1952630cbd1785",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.03285497426986694,
    "fitness": 0.8322890252536799,
    "legacy_fitness": 0.8322890252536799,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7100715866840618,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8363044261932373,
        "normalized": 0.16439289419288416,
        "contribution": 0.04109822354822104
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.525838851928711,
        "normalized": 0.7984049796580064,
        "contribution": 0.11976074694870095
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7271077632904053,
        "normalized": 0.6017241918561651,
        "contribution": 0.06017241918561651
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5791411afc05ac8d1884b31c800f18aff195f172eba7b36d8ae0f31cf5deee16",
    "id": "4257e2f84215cf7669ae75e34ea9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.823258876800537,
    "wer": 0.0,
    "identity": 0.7465909719467163,
    "voiceclap": 0.05411580950021744,
    "fitness": 0.7943855048032745,
    "legacy_fitness": 0.7943855048032745,
    "target_cdf": 0.9511644047125467,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.932128429412842,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7017461888822544,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.932128429412842,
        "normalized": 0.9511644047125467,
        "contribution": 0.47558220235627335
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7838149070739746,
        "normalized": 0.8559956150601711,
        "contribution": 0.12839934225902566
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.975736379623413,
        "normalized": 0.6838583782777219,
        "contribution": 0.06838583782777219
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a112acab475f744da9981a58b4391c81958ddcb0a5a7c80623afdc3c977c18e1",
    "id": "426e1cb3479bc91c2518a4040dab",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.400654315948486,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.045018892735242844,
    "fitness": 0.8307455103952948,
    "legacy_fitness": 0.8307455103952948,
    "target_cdf": 0.9721939248560396,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1999640464782715,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7135559194758089,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1999640464782715,
        "normalized": 0.9721939248560396,
        "contribution": 0.4860969624280198
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9041526317596436,
        "normalized": 0.18447617999078308,
        "contribution": 0.04611904499769577
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5436301231384277,
        "normalized": 0.8027714861707109,
        "contribution": 0.12041572292560664
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.749110221862793,
        "normalized": 0.6092418912448665,
        "contribution": 0.060924189124486654
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d22196ed65d2bb00fe7b3fd4319fd8e4f173495790feff5d36fbfeca9e56c3cb",
    "id": "426ffaf89ccaf9ef0ef69fe18444",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 5.982900142669678,
    "genuineness": 3.9676074981689453,
    "wer": 0.06666666666666667,
    "identity": 0.10688157379627228,
    "voiceclap": -0.04000640660524368,
    "fitness": 0.7550861681726022,
    "legacy_fitness": 0.7550861681726022,
    "target_cdf": 0.988262823627061,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5651941299438477,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7958286409895285,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5651941299438477,
        "normalized": 0.988262823627061,
        "contribution": 0.4941314118135305
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1076972484588623,
        "normalized": 0.25332518159841916,
        "contribution": 0.06333129539960479
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.646700859069824,
        "normalized": 0.9656443012467251,
        "contribution": 0.14484664518700877
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.144070625305176,
        "normalized": 0.9351928858938454,
        "contribution": 0.09351928858938455
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0678e5c219ef1aae32bb46738aa1abb20289e4917ca44aa486a718429e13f045",
    "id": "4276e110b870fe7c3640336161f7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.005255388095974922,
    "fitness": 0.8346856814602706,
    "legacy_fitness": 0.8346856814602706,
    "target_cdf": 0.9967235449909552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.035935401916504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.764760472265054,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.035935401916504,
        "normalized": 0.9967235449909552,
        "contribution": 0.4983617724954776
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9245071411132812,
        "normalized": 0.1907869738584032,
        "contribution": 0.0476967434646008
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.426788330078125,
        "normalized": 0.8102739668234495,
        "contribution": 0.08102739668234496
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e3c7d5d32d0ba37341523327c731bc0d46dd0ee69381ebbbb5e12081fed9204a",
    "id": "42aeee36b36a028aeb82525b75e7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.099404335021973,
    "wer": 0.0,
    "identity": 0.5756174325942993,
    "voiceclap": -0.01048365980386734,
    "fitness": 0.8282073795056277,
    "legacy_fitness": 0.8282073795056277,
    "target_cdf": 0.9997673708708951,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.84755802154541,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7618933800256343,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.84755802154541,
        "normalized": 0.9997673708708951,
        "contribution": 0.49988368543544753
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5716965198516846,
        "normalized": 0.1000529371132764,
        "contribution": 0.0250132342783191
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.9580512046813965,
        "normalized": 0.9818588747524496,
        "contribution": 0.14727883121286744
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.862438201904297,
        "normalized": 0.897176290990003,
        "contribution": 0.0897176290990003
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d9ee0f7aacc275bcfe29a2a6c37dcc2e1f8103c6cdf879ae0370b3b6511e2aac",
    "id": "42c83c5d18631eaef71f28032711",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.3881025314331055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.110545314848423,
    "fitness": 0.8251088764705328,
    "legacy_fitness": 0.8251088764705328,
    "target_cdf": 0.9726892523841197,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2080800533294678,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7240271957486691,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2080800533294678,
        "normalized": 0.9726892523841197,
        "contribution": 0.48634462619205987
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7548866271972656,
        "normalized": 0.14223933958928087,
        "contribution": 0.03555983489732022
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6726183891296387,
        "normalized": 0.8326884740215883,
        "contribution": 0.12490327110323823
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2771711349487305,
        "normalized": 0.7721946355605077,
        "contribution": 0.07721946355605078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1dd638157816152a5e7a0d2f2e0ee01b07d8fbdb9531bc028ebd24ef341bc46f",
    "id": "42d2bad87716de42be32b1ced3a5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": 0.44028693437576294,
    "voiceclap": 0.08130195736885071,
    "fitness": 0.8292813166466498,
    "legacy_fitness": 0.8292813166466498,
    "target_cdf": 0.9867481280341037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5164966583251953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7883990475240613,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5164966583251953,
        "normalized": 0.9867481280341037,
        "contribution": 0.49337406401705186
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.161975860595703,
        "normalized": 0.27373275044956896,
        "contribution": 0.06843318761239224
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.068478107452393,
        "normalized": 0.9053242349151492,
        "contribution": 0.13579863523727237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9328465461730957,
        "normalized": 0.9079316065734483,
        "contribution": 0.09079316065734483
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c965e97a3c1fa5d6f7fb14ada0d8b71aaa2637eb8fc7b2fe272d7c40efa8648e",
    "id": "4307c76a099ee9e5dacf195e33ec",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.038216762244701385,
    "fitness": 0.7640801469558969,
    "legacy_fitness": 0.7640801469558969,
    "target_cdf": 0.8577606996048479,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.323411464691162,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7003817470220269,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.323411464691162,
        "normalized": 0.8577606996048479,
        "contribution": 0.42888034980242395
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.338381052017212,
        "normalized": 0.34515387848215506,
        "contribution": 0.08628846962053877
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.174457550048828,
        "normalized": 0.7010225607218077,
        "contribution": 0.10515338410827114
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.387183666229248,
        "normalized": 0.800595434907931,
        "contribution": 0.0800595434907931
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "346cc5a8a7c1c8e4baa38da05a47c9629b9fb5f30e008749e6995f5926e9abc9",
    "id": "43c6d4c75c9634620046d495fb05",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.20609712600708,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.007178478874266148,
    "fitness": 0.8030266427882493,
    "legacy_fitness": 0.8030266427882493,
    "target_cdf": 0.9321750211493136,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.761687755584717,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7446506289841269,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.761687755584717,
        "normalized": 0.9321750211493136,
        "contribution": 0.4660875105746568
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3010644912719727,
        "normalized": 0.32945945286060696,
        "contribution": 0.08236486321515174
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4368815422058105,
        "normalized": 0.7757183877941347,
        "contribution": 0.1163577581691202
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.378382682800293,
        "normalized": 0.7984049702519815,
        "contribution": 0.07984049702519816
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f767a69a7ade1f8f29f514aec986ddd61e39d40f4a9b1d483dbb1b9cc116c318",
    "id": "442c99356d068adbc023420f0b49",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": 0.6268970966339111,
    "voiceclap": -0.0840425193309784,
    "fitness": 0.8266445278791565,
    "legacy_fitness": 0.8266445278791565,
    "target_cdf": 0.961270594530896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.045755386352539,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.669379931824798,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.045755386352539,
        "normalized": 0.961270594530896,
        "contribution": 0.480635297265448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.290102005004883,
        "normalized": 0.7352947867084237,
        "contribution": 0.11029421800626356
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4319987297058105,
        "normalized": 0.4983442069459631,
        "contribution": 0.049834420694596315
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "82f1bf403aa1be5f9ecfcd31a9310589869ec4224d6bdfa61e583f0ac87963ba",
    "id": "4474efb276ebad003f5ffbc5c01f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.157639026641846,
    "genuineness": 3.8295347690582275,
    "wer": 0.0,
    "identity": 0.48993682861328125,
    "voiceclap": 0.07038655132055283,
    "fitness": 0.812390273582418,
    "legacy_fitness": 0.812390273582418,
    "target_cdf": 0.9977706984112247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.165794849395752,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8361536758627419,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.165794849395752,
        "normalized": 0.9977706984112247,
        "contribution": 0.49888534920561234
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3858747482299805,
        "normalized": 0.3655171567228084,
        "contribution": 0.0913792891807021
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.064799785614014,
        "normalized": 0.9856469755282963,
        "contribution": 0.14784704632924445
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.760141372680664,
        "normalized": 0.9804199114718302,
        "contribution": 0.09804199114718302
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a5b674ade8ed702d347300e5b11b9efbc76159c1f07e293345ea933b2c379ff0",
    "id": "44ba4633251f5f7971d4448cd62c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.595211505889893,
    "wer": 0.0,
    "identity": 0.43829381465911865,
    "voiceclap": -0.043095409870147705,
    "fitness": 0.8119720466578892,
    "legacy_fitness": 0.8119720466578892,
    "target_cdf": 0.9235966379731679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6967577934265137,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6430137088139937,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6967577934265137,
        "normalized": 0.9235966379731679,
        "contribution": 0.46179831898658397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.9676318168640137,
        "normalized": 0.6352170794829098,
        "contribution": 0.09528256192243648
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4839797019958496,
        "normalized": 0.5167475031342208,
        "contribution": 0.051674750313422085
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "61a2f0d6cfe88699e82ece81d8526676bb4e3119f95caa6597bd1407245743bd",
    "id": "453f32f18d026babbb46cdb887b4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": 0.6479377150535583,
    "voiceclap": 0.01876933127641678,
    "fitness": 0.8000549862617008,
    "legacy_fitness": 0.8000549862617008,
    "target_cdf": 0.9409148593413338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.834733486175537,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6835234231370629,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.834733486175537,
        "normalized": 0.9409148593413338,
        "contribution": 0.4704574296706669
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5436301231384277,
        "normalized": 0.8027714861707109,
        "contribution": 0.12041572292560664
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8415210247039795,
        "normalized": 0.640342746822993,
        "contribution": 0.0640342746822993
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5b39e371f85e563852f13d9db9d1c5f9ad9025280b75ac630d0285935f08b9e3",
    "id": "45482a0ac3be2bd371f5f08757ef",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": 0.4621681869029999,
    "voiceclap": 0.005761997774243355,
    "fitness": 0.8206299058026378,
    "legacy_fitness": 0.8206299058026378,
    "target_cdf": 0.997542097454529,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.13332986831665,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8257282413375886,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.13332986831665,
        "normalized": 0.997542097454529,
        "contribution": 0.4987710487272645
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3248114585876465,
        "normalized": 0.33941420432120517,
        "contribution": 0.08485355108030129
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.771240711212158,
        "normalized": 0.9731772088639971,
        "contribution": 0.14597658132959956
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.425703048706055,
        "normalized": 0.9612706020042315,
        "contribution": 0.09612706020042316
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a2da0a5003ef2589d23189eb853ab53c038f6c5872047be4748aee5137bddcc2",
    "id": "456907f9238fdc51f46b7869417e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.149612903594971,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.015617337077856064,
    "fitness": 0.8007539550949735,
    "legacy_fitness": 0.8007539550949735,
    "target_cdf": 0.9311452866573339,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7535715103149414,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7116920669119643,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7535715103149414,
        "normalized": 0.9311452866573339,
        "contribution": 0.46557264332866694
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8770133256912231,
        "normalized": 0.17626644875656883,
        "contribution": 0.04406661218914221
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.730440616607666,
        "normalized": 0.845096013305189,
        "contribution": 0.12676440199577835
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2067627906799316,
        "normalized": 0.7528840939837675,
        "contribution": 0.07528840939837676
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ddd092afc89510765b989618ab5f0c591722c8d98e287977f5dba342cf01594f",
    "id": "458f0d5b4625d344609420a5446e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.8546390533447266,
    "wer": 0.0,
    "identity": 0.5964784622192383,
    "voiceclap": 0.0056278035044670105,
    "fitness": 0.7812754176606014,
    "legacy_fitness": 0.7812754176606014,
    "target_cdf": 0.9247120186784414,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.704874038696289,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7126963818845455,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.704874038696289,
        "normalized": 0.9247120186784414,
        "contribution": 0.4623560093392207
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8600512742996216,
        "normalized": 0.17125467685227086,
        "contribution": 0.042813669213067715
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.290372610092163,
        "normalized": 0.7757183970251897,
        "contribution": 0.07757183970251898
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "10f75a420894e7ad05629e1dda9e862af585cf1663c4bf9f60269adcbcb1f835",
    "id": "45f2290765d2a5386715b283bd88",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.11530134081840515,
    "fitness": 0.8063129486187728,
    "legacy_fitness": 0.8063129486187728,
    "target_cdf": 0.9662346008508267,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.110685348510742,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7905056436140245,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.110685348510742,
        "normalized": 0.9662346008508267,
        "contribution": 0.48311730042541334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.390963315963745,
        "normalized": 0.3677227903226994,
        "contribution": 0.09193069758067485
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.872772216796875,
        "normalized": 0.8729869940654369,
        "contribution": 0.13094804910981553
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.580805778503418,
        "normalized": 0.8450959649812083,
        "contribution": 0.08450959649812084
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3577b8711be11316b82e76368f2e3409b660b9d15e0a708cedfef9f9cee349ed",
    "id": "467b0685e6cf2db130e66156701b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 4.143336772918701,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06165201589465141,
    "fitness": 0.816599030051122,
    "legacy_fitness": 0.816599030051122,
    "target_cdf": 0.962564179804901,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.06198787689209,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7469799030318779,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.06198787689209,
        "normalized": 0.962564179804901,
        "contribution": 0.4812820899024505
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8363044261932373,
        "normalized": 0.16439289419288416,
        "contribution": 0.04109822354822104
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.246392726898193,
        "normalized": 0.9290496824330181,
        "contribution": 0.1393574523649527
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6160099506378174,
        "normalized": 0.852421372162536,
        "contribution": 0.08524213721625361
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9b8fd9f3ee85b031f552503e4041c8370f879166976eb7f506c8d6ad0f2b6c13",
    "id": "469b216f3822a246842bab178e41",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.055472373962402,
    "wer": 0.0,
    "identity": 0.6558430790901184,
    "voiceclap": -0.0928552895784378,
    "fitness": 0.7626545897099276,
    "legacy_fitness": 0.7626545897099276,
    "target_cdf": 0.8663657806263168,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.36399245262146,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6439953691779411,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.36399245262146,
        "normalized": 0.8663657806263168,
        "contribution": 0.4331828903131584
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3568201065063477,
        "normalized": 0.7541161908392326,
        "contribution": 0.11311742862588489
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8107173442840576,
        "normalized": 0.6300671571887225,
        "contribution": 0.06300671571887226
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "37899d62a5c7f01af4b449d298900f716e97496d64258446f2e97fae00f133b0",
    "id": "46db29d9a1d5dd57df3edf5b12bd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 3.7667744159698486,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04916573315858841,
    "fitness": 0.8153599725747196,
    "legacy_fitness": 0.8153599725747196,
    "target_cdf": 0.9990117801396228,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.425514221191406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7930451149855005,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.425514221191406,
        "normalized": 0.9990117801396228,
        "contribution": 0.4995058900698114
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.982177972793579,
        "normalized": 0.2093744108235418,
        "contribution": 0.05234360270588545
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.993634223937988,
        "normalized": 0.9832067001034072,
        "contribution": 0.14748100501551106
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.161672592163086,
        "normalized": 0.9371461719429255,
        "contribution": 0.09371461719429255
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b3edb330fcc846f1b084f02f2016ea37f55f437010b666c8836037aaef0ae840",
    "id": "46e68a0ab828ed1b021989c8f826",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 3.9425034523010254,
    "wer": 0.0,
    "identity": 0.5538346171379089,
    "voiceclap": 0.11654932051897049,
    "fitness": 0.8192971478709903,
    "legacy_fitness": 0.8192971478709903,
    "target_cdf": 0.9993904532610031,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.571606159210205,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8293565643281704,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.571606159210205,
        "normalized": 0.9993904532610031,
        "contribution": 0.49969522663050153
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3248114585876465,
        "normalized": 0.33941420432120517,
        "contribution": 0.08485355108030129
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.1003828048706055,
        "normalized": 0.986748131774742,
        "contribution": 0.1480122197662113
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.522514343261719,
        "normalized": 0.9679556685115622,
        "contribution": 0.09679556685115623
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "54ec48633b2bf460b08c1ce6160cc8c324e84310cf8bc9658628c6faed4454ab",
    "id": "474a2b676f354cc400c4e72f909a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.457139015197754,
    "wer": 0.06666666666666667,
    "identity": 0.5480867624282837,
    "voiceclap": -0.1624757945537567,
    "fitness": 0.7443444679604189,
    "legacy_fitness": 0.7443444679604189,
    "target_cdf": 0.911742319044517,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6155953407287598,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6695230958257369,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6155953407287598,
        "normalized": 0.911742319044517,
        "contribution": 0.4558711595222585
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9312918186187744,
        "normalized": 0.19291961867711876,
        "contribution": 0.04822990466927969
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2678627967834473,
        "normalized": 0.7288620844464377,
        "contribution": 0.10932931266696566
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6093943119049072,
        "normalized": 0.5609271896723302,
        "contribution": 0.05609271896723303
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "61710050b4820f834c911e9cc789cc9edaa97ed3fc845321e7456fa179f7520c",
    "id": "4750292b38090958b6f4aacaf422",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.432034492492676,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.2096492052078247,
    "fitness": 0.7201742022260543,
    "legacy_fitness": 0.7201742022260543,
    "target_cdf": 0.861246668375216,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.339643716812134,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6240640684437933,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.339643716812134,
        "normalized": 0.861246668375216,
        "contribution": 0.430623334187608
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9143297672271729,
        "normalized": 0.18761514602105922,
        "contribution": 0.046903786505264805
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0165584087371826,
        "normalized": 0.6512405794126703,
        "contribution": 0.09768608691190055
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4042205810546875,
        "normalized": 0.4885086083901984,
        "contribution": 0.04885086083901984
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "dd6e99177296a841b4ad24767639149612d0d211ab5ffa271af404ea96f04b3e",
    "id": "48335e0fbc82c51f3540229f21b9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.532451152801514,
    "wer": 0.0,
    "identity": 0.6067054867744446,
    "voiceclap": -0.08870622515678406,
    "fitness": 0.8173846661646232,
    "legacy_fitness": 0.8173846661646232,
    "target_cdf": 0.9436232446827622,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8590822219848633,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6560017957300445,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8590822219848633,
        "normalized": 0.9436232446827622,
        "contribution": 0.4718116223413811
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.693823218345642,
        "normalized": 0.127012998991818,
        "contribution": 0.0317532497479545
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1522183418273926,
        "normalized": 0.6942076233845773,
        "contribution": 0.1041311435076866
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.3888187408447266,
        "normalized": 0.48305780133022336,
        "contribution": 0.048305780133022336
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c0f2e7eff1d67a679de56e7fe95fa5f603e86cd78b105ce56059e6c028b606c3",
    "id": "48685cb423f6e2f6f807927c08db",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.610371112823486,
    "genuineness": 4.262581825256348,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.025034241378307343,
    "fitness": 0.83510082097909,
    "legacy_fitness": 0.83510082097909,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7143573521250154,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5649117231369019,
        "normalized": 0.0986881865760929,
        "contribution": 0.024672046644023227
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8016064167022705,
        "normalized": 0.8595110162463788,
        "contribution": 0.1289266524369568
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9801368713378906,
        "normalized": 0.6852468593788036,
        "contribution": 0.06852468593788036
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ad77186d098ea8b88f420641bb46005698352645b738778623752fb559e01427",
    "id": "487ba2a5f063fb5e79426c5ee079",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.042832277715206146,
    "fitness": 0.8160834580956967,
    "legacy_fitness": 0.8160834580956967,
    "target_cdf": 0.9972925007937494,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.100865364074707,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7932933304746047,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.100865364074707,
        "normalized": 0.9972925007937494,
        "contribution": 0.4986462503968747
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.033064126968384,
        "normalized": 0.2266273368759334,
        "contribution": 0.05665683421898335
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.8157196044921875,
        "normalized": 0.9755091035016348,
        "contribution": 0.14632636552524522
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9944534301757812,
        "normalized": 0.9166388033350146,
        "contribution": 0.09166388033350147
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1ff0f9fe1f6370461fb267e5dd3771f0b393487d8ac173031eb10a4be8be9371",
    "id": "489bbc3515b5c81d6b3c9a833a28",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.05190243199467659,
    "fitness": 0.8293236569224475,
    "legacy_fitness": 0.8293236569224475,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7649114742649434,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.961823582649231,
        "normalized": 0.2026956548789064,
        "contribution": 0.0506739137197266
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.952833652496338,
        "normalized": 0.8870402984992682,
        "contribution": 0.13305604477489022
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5984079837799072,
        "normalized": 0.8487882013251657,
        "contribution": 0.08487882013251657
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9a3a8a537a559b9d4d4c3e4aac6511f119c30cd6e74954da6b098eaf2e2f1b32",
    "id": "48bd867f2b88dfe6347c44ea7b5e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.76,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.6663577556610107,
    "wer": 0.0,
    "identity": 0.4767985939979553,
    "voiceclap": -0.009392019361257553,
    "fitness": 0.8068312093106056,
    "legacy_fitness": 0.8068312093106056,
    "target_cdf": 0.9867481280341037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5164966583251953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7457337110110636,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5164966583251953,
        "normalized": 0.9867481280341037,
        "contribution": 0.49337406401705186
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5716965198516846,
        "normalized": 0.1000529371132764,
        "contribution": 0.0250132342783191
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.48657751083374,
        "normalized": 0.9534909337644679,
        "contribution": 0.1430236400646702
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.572004795074463,
        "normalized": 0.8432277265102239,
        "contribution": 0.08432277265102239
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d40b71bcd1edb64b4ba81a54d1e34f9be7b8f3629a392fd619f47586807b79aa",
    "id": "48ec63b465b7726e7c4ce0420632",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.369274139404297,
    "wer": 0.06666666666666667,
    "identity": 0.25797030329704285,
    "voiceclap": -0.13085684180259705,
    "fitness": 0.7616704108993004,
    "legacy_fitness": 0.7616704108993004,
    "target_cdf": 0.9585742155738795,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0132906436920166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7366619069446578,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0132906436920166,
        "normalized": 0.9585742155738795,
        "contribution": 0.4792871077869397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9211146831512451,
        "normalized": 0.1897260605487444,
        "contribution": 0.0474315151371861
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3079745769500732,
        "normalized": 0.780368248743694,
        "contribution": 0.07803682487436941
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c3a576eab87ad1d54f43f714c4d092169db0d310164f116d0c959cc073b7a783",
    "id": "493739ef0d48a2ca5d5cb39b8597",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": 0.7228586673736572,
    "voiceclap": -0.020606711506843567,
    "fitness": 0.7872730191831876,
    "legacy_fitness": 0.7872730191831876,
    "target_cdf": 0.9427315386702796,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.850965976715088,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7219870056877506,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.850965976715088,
        "normalized": 0.9427315386702796,
        "contribution": 0.4713657693351398
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8600512742996216,
        "normalized": 0.17125467685227086,
        "contribution": 0.042813669213067715
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2287654876708984,
        "normalized": 0.7590110799338052,
        "contribution": 0.07590110799338053
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "785e509b0bc7232354aff494862eebf659a457bbfc7c9101753f20927aeb2ed9",
    "id": "495035a98b43d80e015992238fec",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 3.9801597595214844,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.06628823280334473,
    "fitness": 0.8093448142846237,
    "legacy_fitness": 0.8093448142846237,
    "target_cdf": 0.9668165486335377,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1188015937805176,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7496935062941724,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1188015937805176,
        "normalized": 0.9668165486335377,
        "contribution": 0.48340827431676886
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0161020755767822,
        "normalized": 0.22078898327927737,
        "contribution": 0.05519724581981934
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4311888217926025,
        "normalized": 0.8113312252784615,
        "contribution": 0.08113312252784616
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "125d8d9812ffd7aa014f1fbede532650031982266a7221fb3c1dd04de1bb9b64",
    "id": "49aba1ccd06b40596c52438ccc05",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.2182147204875946,
    "fitness": 0.7849975724121665,
    "legacy_fitness": 0.7849975724121665,
    "target_cdf": 0.8914684095185468,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.493852138519287,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6375055570188679,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.493852138519287,
        "normalized": 0.8914684095185468,
        "contribution": 0.4457342047592734
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.632759928703308,
        "normalized": 0.11295969157872005,
        "contribution": 0.02823992289468001
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1700096130371094,
        "normalized": 0.6996650705664804,
        "contribution": 0.10494976058497206
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6809024810791016,
        "normalized": 0.5858166877994249,
        "contribution": 0.05858166877994249
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "336be2d1b14acda678941036f8c3aff34dd19a2db065517629dc71be9d10af02",
    "id": "49b835af2156b28fa466d5853807",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": 0.42376312613487244,
    "voiceclap": 0.0061853667721152306,
    "fitness": 0.8347446922137605,
    "legacy_fitness": 0.8347446922137605,
    "target_cdf": 0.9919413812308048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7112860679626465,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7709842739936159,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7112860679626465,
        "normalized": 0.9919413812308048,
        "contribution": 0.4959706906154024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.005924940109253,
        "normalized": 0.21732775018194794,
        "contribution": 0.054331937545486986
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.139644145965576,
        "normalized": 0.9154342857630313,
        "contribution": 0.1373151428644547
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5279998779296875,
        "normalized": 0.8336650296827179,
        "contribution": 0.0833665029682718
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "109f52000aaa5956f061718313c5550ba469e0156c27c73eefa0eb23fd426b87",
    "id": "49d28370d86f9ddf04c64ba40b87",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": 0.39042729139328003,
    "voiceclap": 0.00028959475457668304,
    "fitness": 0.8172659311717406,
    "legacy_fitness": 0.8172659311717406,
    "target_cdf": 0.9974200236848305,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1170973777771,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7513718906417758,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1170973777771,
        "normalized": 0.9974200236848305,
        "contribution": 0.49871001184241526
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6259750127792358,
        "normalized": 0.11146953771947371,
        "contribution": 0.027867384429868428
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.353141784667969,
        "normalized": 0.9409148986309699,
        "contribution": 0.1411372347946455
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.54120135307312,
        "normalized": 0.8365725957484663,
        "contribution": 0.08365725957484664
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "489e068646eeeb6ba2d6bf9ffc7f0c49213f60db05d69b472df15dfcb198fe22",
    "id": "49d449fab1b2474f624a99cb3ad8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.44,
    "cfg": 3,
    "generation": null,
    "ce": 6.483288764953613,
    "genuineness": 3.961331605911255,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09070010483264923,
    "fitness": 0.7966643577780742,
    "legacy_fitness": 0.7966643577780742,
    "target_cdf": 0.9390533101491965,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8185012340545654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7407794627360876,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8185012340545654,
        "normalized": 0.9390533101491965,
        "contribution": 0.46952665507459823
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.16536808013916,
        "normalized": 0.2750346407474058,
        "contribution": 0.06875866018685145
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.694857597351074,
        "normalized": 0.8375344060053678,
        "contribution": 0.12563016090080517
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2639694213867188,
        "normalized": 0.7686398657383282,
        "contribution": 0.07686398657383282
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "90e0ffc741d0b0f0dc571d481bb4d0a2c65251be3834a456179056e293a2874d",
    "id": "49dada128e02a147c1da25efc0e8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.8420867919921875,
    "wer": 0.0,
    "identity": 0.5944943428039551,
    "voiceclap": 0.07271759957075119,
    "fitness": 0.8126131618422119,
    "legacy_fitness": 0.8126131618422119,
    "target_cdf": 0.9896249303026692,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6138916015625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7365304560216095,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6138916015625,
        "normalized": 0.9896249303026692,
        "contribution": 0.4948124651513346
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5581269264221191,
        "normalized": 0.09733711108070414,
        "contribution": 0.024334277770176035
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.344245910644531,
        "normalized": 0.9399897500962073,
        "contribution": 0.1409984625144311
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2463674545288086,
        "normalized": 0.7638525058566775,
        "contribution": 0.07638525058566775
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d78a354d17ca5a661c01f6d27d816e4a67c7d45008502d8cde35fc2cdf3bd555",
    "id": "49e9e354b518534b3edd035d7b3c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.080576419830322,
    "wer": 0.0,
    "identity": 0.44655391573905945,
    "voiceclap": 0.05922623723745346,
    "fitness": 0.8117486873932658,
    "legacy_fitness": 0.8117486873932658,
    "target_cdf": 0.961270594530896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.045755386352539,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7721385532420715,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.045755386352539,
        "normalized": 0.961270594530896,
        "contribution": 0.480635297265448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.216254234313965,
        "normalized": 0.2949218491618698,
        "contribution": 0.07373046229046745
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.483994960784912,
        "normalized": 0.8237335781548634,
        "contribution": 0.08237335781548634
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "252a7e85ad5c3e22f6ff79d84ac11b4cf6ac9eb491404c5d6c3068ce5081c33a",
    "id": "4a8ef2c1a3872c08d19fa0746d63",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.5786004066467285,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": 0.5175176858901978,
    "voiceclap": -0.06690948456525803,
    "fitness": 0.7647622928632505,
    "legacy_fitness": 0.7647622928632505,
    "target_cdf": 0.8375344088898514,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.234132766723633,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6260069616852069,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.234132766723633,
        "normalized": 0.8375344088898514,
        "contribution": 0.4187672044449257
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8837981224060059,
        "normalized": 0.17829687836447006,
        "contribution": 0.044574219591117514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0165584087371826,
        "normalized": 0.6512405794126703,
        "contribution": 0.09768608691190055
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.870124340057373,
        "normalized": 0.6497945073726306,
        "contribution": 0.06497945073726306
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b4866c53d269730a9df1c3d67c7dbb4d51f146eec9ee7247091f369c57dd244a",
    "id": "4af365b1f2dfc930dbeb17348206",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 3.8044304847717285,
    "wer": 0.0,
    "identity": 0.5885512232780457,
    "voiceclap": 0.03578871861100197,
    "fitness": 0.8151523323731584,
    "legacy_fitness": 0.8151523323731584,
    "target_cdf": 0.9998269661023145,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.928720474243164,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7932039108914646,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.928720474243164,
        "normalized": 0.9998269661023145,
        "contribution": 0.49991348305115724
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9720008373260498,
        "normalized": 0.206019001985337,
        "contribution": 0.05150475049633425
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.2427144050598145,
        "normalized": 0.9904545174589181,
        "contribution": 0.1485681776188377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.1176676750183105,
        "normalized": 0.9321749972513534,
        "contribution": 0.09321749972513535
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "266c84ef58f415386dd93195191a8d001f3f1d77761454768cb234909f5f1f04",
    "id": "4b00f590f0a2f06853657dc67ad8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.722842216491699,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.15642613172531128,
    "fitness": 0.8119812880099322,
    "legacy_fitness": 0.8119812880099322,
    "target_cdf": 0.9998867659053218,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.042347431182861,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8612392212638698,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.042347431182861,
        "normalized": 0.9998867659053218,
        "contribution": 0.4999433829526609
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5936598777770996,
        "normalized": 0.4583894155259455,
        "contribution": 0.11459735388148637
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.456212043762207,
        "normalized": 0.9943171611150623,
        "contribution": 0.14914757416725935
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.654529571533203,
        "normalized": 0.9755091026246321,
        "contribution": 0.09755091026246321
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "24ad304f411749cdf22906f9361313bc466ad5aca3db11e023249a448162b80a",
    "id": "4bd27cb84dc4ee7b242671418919",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.009762164205312729,
    "fitness": 0.8304383425747524,
    "legacy_fitness": 0.8304383425747524,
    "target_cdf": 0.9763929662774343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.273009777069092,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7389081100374956,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.273009777069092,
        "normalized": 0.9763929662774343,
        "contribution": 0.48819648313871716
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8193423748016357,
        "normalized": 0.1596022950986577,
        "contribution": 0.039900573774664425
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2111635208129883,
        "normalized": 0.754116172534443,
        "contribution": 0.07541161725344431
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "341a0f63ff500c6aa2bb3723581a27c100b3ff4dafeb32921866f06664f847da",
    "id": "4c3382b30531c4c8c65b729634a0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": 0.5717635154724121,
    "voiceclap": -0.0010627172887325287,
    "fitness": 0.8007159753956055,
    "legacy_fitness": 0.8007159753956055,
    "target_cdf": 0.9542463867500731,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9645934104919434,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7165556446258902,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9645934104919434,
        "normalized": 0.9542463867500731,
        "contribution": 0.47712319337503656
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.997312545776367,
        "normalized": 0.8943502454552662,
        "contribution": 0.13415253681828992
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2287654876708984,
        "normalized": 0.7590110799338052,
        "contribution": 0.07590110799338053
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ea3091757dea7bd7eb1f07479909e14f362d7be1e21e56dcaba79a72d96ab5ef",
    "id": "4c92402a525f3e09d5919af86bef",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.532451152801514,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.02165926992893219,
    "fitness": 0.828164686083702,
    "legacy_fitness": 0.828164686083702,
    "target_cdf": 0.9631976153848724,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0701041221618652,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7023099638757078,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0701041221618652,
        "normalized": 0.9631976153848724,
        "contribution": 0.4815988076924362
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3701634407043457,
        "normalized": 0.7577923025461362,
        "contribution": 0.11366884538192042
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9735360145568848,
        "normalized": 0.6831631248576189,
        "contribution": 0.0683163124857619
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "70bcafa5d7a0414b602c7331d47f7f949698d26c1c28c2cab3a0fe9f5602417d",
    "id": "4c95bd28a75af26ddc6caa61e427",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": 0.7031067609786987,
    "voiceclap": 0.05102098733186722,
    "fitness": 0.8075276460989883,
    "legacy_fitness": 0.8075276460989883,
    "target_cdf": 0.9696036431679269,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1593828201293945,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7304668234061139,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1593828201293945,
        "normalized": 0.9696036431679269,
        "contribution": 0.48480182158396345
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.970625400543213,
        "normalized": 0.8900064909716228,
        "contribution": 0.1335009736457434
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3519797325134277,
        "normalized": 0.7917476313694349,
        "contribution": 0.0791747631369435
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6676efa6aca9410acd913318c826efa95f98667a338fc3ec345bf9faeb0cb443",
    "id": "4cb5f71ec93ddc036f983b825354",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.1166682243347168,
    "fitness": 0.8221896552713251,
    "legacy_fitness": 0.8221896552713251,
    "target_cdf": 0.9987822818103504,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.360584259033203,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8579337980633573,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.360584259033203,
        "normalized": 0.9987822818103504,
        "contribution": 0.4993911409051752
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.577970027923584,
        "normalized": 0.45122837352159934,
        "contribution": 0.11280709338039983
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.064799785614014,
        "normalized": 0.9856469755282963,
        "contribution": 0.14784704632924445
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.724937438964844,
        "normalized": 0.9788851744853782,
        "contribution": 0.09788851744853783
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "89cb9787fbf2c05b69e3a0fa549f52e6cbe08e7829087fd1fc06af27183bab28",
    "id": "4d450ba79f7ca452dae75487137d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 3.886019229888916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0444306805729866,
    "fitness": 0.8102710815013567,
    "legacy_fitness": 0.8102710815013567,
    "target_cdf": 0.9741313242397945,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.232429027557373,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.764044436761373,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.232429027557373,
        "normalized": 0.9741313242397945,
        "contribution": 0.48706566211989727
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.03645658493042,
        "normalized": 0.2278053936557905,
        "contribution": 0.056951348413947625
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.0239996910095215,
        "normalized": 0.8985685934084969,
        "contribution": 0.13478528901127454
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6160099506378174,
        "normalized": 0.852421372162536,
        "contribution": 0.08524213721625361
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a246196d05c7827d2eac58ce35c488ae69587ef756c7d57f247d309854d7d3f6",
    "id": "4d9067622f453333822ee8092bea",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": 0.39478689432144165,
    "voiceclap": 0.00187673419713974,
    "fitness": 0.7973298550287693,
    "legacy_fitness": 0.7973298550287693,
    "target_cdf": 0.9668165486335377,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1188015937805176,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7923729318630307,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1188015937805176,
        "normalized": 0.9668165486335377,
        "contribution": 0.48340827431676886
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2298238277435303,
        "normalized": 0.3003348835264067,
        "contribution": 0.07508372088160167
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.3709330558776855,
        "normalized": 0.9427315543087618,
        "contribution": 0.14140973314631425
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.056060791015625,
        "normalized": 0.9247120351834592,
        "contribution": 0.09247120351834592
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2f604cc41cf5cc10f8811d939bb71577b38d9f15030d54ab08fb6f8748d71f0b",
    "id": "4d964a79995563708e79b08a0077",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 4.268857479095459,
    "wer": 0.0,
    "identity": 0.6306183338165283,
    "voiceclap": -0.06124695762991905,
    "fitness": 0.8329888134805375,
    "legacy_fitness": 0.8329888134805375,
    "target_cdf": 0.9964005449628859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.003470420837402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.731062290898085,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.003470420837402,
        "normalized": 0.9964005449628859,
        "contribution": 0.49820027248144294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.436000108718872,
        "normalized": 0.07528798214951904,
        "contribution": 0.01882199553737976
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.379828929901123,
        "normalized": 0.9436232726705875,
        "contribution": 0.14154349090058813
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1099517345428467,
        "normalized": 0.7249653197867423,
        "contribution": 0.07249653197867423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8399b0d60ad025543dc5af5d398b8fb7740ff53bb6b464e91dfd28085cf54266",
    "id": "4d964f38bc586e5487c0afaccc3b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.143336772918701,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11398961395025253,
    "fitness": 0.7972604366772879,
    "legacy_fitness": 0.7972604366772879,
    "target_cdf": 0.929049708969098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7373390197753906,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6881408234734697,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7373390197753906,
        "normalized": 0.929049708969098,
        "contribution": 0.464524854484549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.508047342300415,
        "normalized": 0.7939810079308474,
        "contribution": 0.1190971511896271
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.021941661834717,
        "normalized": 0.6983048327926806,
        "contribution": 0.06983048327926807
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7b22408fa565a360477b7f9b9a3ecab39aeda728afeabcbb315470a119735662",
    "id": "4dca502fc4876ce557f20b031f53",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.7291181087493896,
    "wer": 0.0,
    "identity": 0.774194598197937,
    "voiceclap": 0.04175751656293869,
    "fitness": 0.794210530075801,
    "legacy_fitness": 0.794210530075801,
    "target_cdf": 0.9638222590155852,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0782203674316406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7080174948907219,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0782203674316406,
        "normalized": 0.9638222590155852,
        "contribution": 0.4819111295077926
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5784813165664673,
        "normalized": 0.10143141696818753,
        "contribution": 0.02535785424204688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8816680908203125,
        "normalized": 0.8746060247794507,
        "contribution": 0.1311909037169176
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0131406784057617,
        "normalized": 0.6955760742396476,
        "contribution": 0.06955760742396476
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9a8e904f8edba29dc5c8c0bf0fd824a434da228ede463a9f79612761a7ba8ded",
    "id": "4ded46bfe0c2d39b351d8812e982",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.594485759735107,
    "genuineness": 4.557555198669434,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09940096735954285,
    "fitness": 0.7809709806672642,
    "legacy_fitness": 0.7809709806672642,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6244658695088309,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8363044261932373,
        "normalized": 0.16439289419288416,
        "contribution": 0.04109822354822104
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1344265937805176,
        "normalized": 0.6887065547816701,
        "contribution": 0.10330598321725051
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5455868244171143,
        "normalized": 0.5385095934736592,
        "contribution": 0.053850959347365925
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a472782a540df8d775001e68c1ec1c7457759bcc88c08903809da3428c15dbb5",
    "id": "4e08c8910baa4e58e68789242205",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": 0.7253511548042297,
    "voiceclap": -0.023013848811388016,
    "fitness": 0.6556869199769046,
    "legacy_fitness": 0.6556869199769046,
    "target_cdf": 0.6605858268134487,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.6416481733322144,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5495803813440132,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.6416481733322144,
        "normalized": 0.6605858268134487,
        "contribution": 0.33029291340672434
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.226431369781494,
        "normalized": 0.29897739616091534,
        "contribution": 0.07474434904022884
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.7385668754577637,
        "normalized": 0.5574595318171561,
        "contribution": 0.08361892977257342
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.749110221862793,
        "normalized": 0.6092418912448665,
        "contribution": 0.060924189124486654
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a3037dc67472773847c29284153bf2161940d5d8104beef3f158bfafaf4c77f6",
    "id": "4e1efcde7ca8d3b4ff6894129fa5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.8,
    "cfg": 1.5,
    "generation": null,
    "ce": 5.911416053771973,
    "genuineness": 4.268857479095459,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09676020592451096,
    "fitness": 0.8004356434561979,
    "legacy_fitness": 0.8004356434561979,
    "target_cdf": 0.9495623609658421,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.915895938873291,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7410232900173993,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.915895938873291,
        "normalized": 0.9495623609658421,
        "contribution": 0.47478118048292106
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1687605381011963,
        "normalized": 0.2763396656306163,
        "contribution": 0.06908491640765407
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6459312438964844,
        "normalized": 0.826751667074274,
        "contribution": 0.1240127500611411
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1319541931152344,
        "normalized": 0.7314444306568307,
        "contribution": 0.07314444306568307
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b5f3214bc9d0e96ca14114ba256b707ca80d384ab7717537c50ada8d90694833",
    "id": "4e58f0bbfbbe8b56b106d92d46ac",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.011539936065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04350532218813896,
    "fitness": 0.7975649700528799,
    "legacy_fitness": 0.7975649700528799,
    "target_cdf": 0.9390533101491965,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8185012340545654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7070420265279169,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8185012340545654,
        "normalized": 0.9390533101491965,
        "contribution": 0.46952665507459823
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.826127290725708,
        "normalized": 0.16150748713874752,
        "contribution": 0.04037687178468688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6726183891296387,
        "normalized": 0.8326884740215883,
        "contribution": 0.12490327110323823
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1011507511138916,
        "normalized": 0.7223522856539355,
        "contribution": 0.07223522856539355
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7b6d8da8f9d94c971ebafd19d2b0617f177b1570952119ecfd06e6c07a6c2ab0",
    "id": "4ee3d1bc97882329e99af519caaa",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 3.7040140628814697,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.11771818995475769,
    "fitness": 0.8098617314273422,
    "legacy_fitness": 0.8098617314273422,
    "target_cdf": 0.9999966023314606,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.886435508728027,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8226325550707222,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.886435508728027,
        "normalized": 0.9999966023314606,
        "contribution": 0.4999983011657303
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.223039150238037,
        "normalized": 0.2976227974003868,
        "contribution": 0.0744056993500967
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 6.078912734985352,
        "normalized": 0.9989584617451555,
        "contribution": 0.1498437692617733
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.848151683807373,
        "normalized": 0.9838478529312191,
        "contribution": 0.09838478529312192
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "415beb5c3879062f813e6cfcc75060527ffaca469067fbecfa4e560fa494e45d",
    "id": "4f75d3c71bfda707065abcbf06f8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.086852550506592,
    "wer": 0.0,
    "identity": 0.6226574778556824,
    "voiceclap": -0.052337028086185455,
    "fitness": 0.8265596093713932,
    "legacy_fitness": 0.8265596093713932,
    "target_cdf": 0.9967235449909552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.035935401916504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8083673082653329,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.035935401916504,
        "normalized": 0.9967235449909552,
        "contribution": 0.4983617724954776
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.195899724960327,
        "normalized": 0.28688766529430837,
        "contribution": 0.07172191632357709
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.5310564041137695,
        "normalized": 0.957170068560992,
        "contribution": 0.14357551028414878
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.25848388671875,
        "normalized": 0.9470810916212937,
        "contribution": 0.09470810916212938
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1b1807b460ba1f10dcb35ce2760c8a2eb59cfd9a594461c1ef327ba57c47b2f1",
    "id": "4fb2a52ae719d7a07b2f11dc7f4d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 3.760498285293579,
    "wer": 0.0,
    "identity": 0.6214464902877808,
    "voiceclap": 0.046810612082481384,
    "fitness": 0.8035435112717464,
    "legacy_fitness": 0.8035435112717464,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7858416553873678,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0873427391052246,
        "normalized": 0.2458838571357379,
        "contribution": 0.061470964283934476
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.477681636810303,
        "normalized": 0.9527255009885056,
        "contribution": 0.14290882514827583
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9152445793151855,
        "normalized": 0.9053242803940382,
        "contribution": 0.09053242803940383
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "16b16bed5ad84dd2fd066ba10c70393828f49ae601648438a89c0f3325161519",
    "id": "512221923cd415be51422f1659d7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.499174118041992,
    "genuineness": 4.21237325668335,
    "wer": 0.0,
    "identity": 0.5624338388442993,
    "voiceclap": -0.123173788189888,
    "fitness": 0.7703319247271787,
    "legacy_fitness": 0.7703319247271787,
    "target_cdf": 0.8646740388286452,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3558762073516846,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6506657439926016,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3558762073516846,
        "normalized": 0.8646740388286452,
        "contribution": 0.4323370194143226
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9075450897216797,
        "normalized": 0.18551887784237608,
        "contribution": 0.04637971946059402
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1032917499542236,
        "normalized": 0.678978341920392,
        "contribution": 0.1018467512880588
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.030742645263672,
        "normalized": 0.7010225382962612,
        "contribution": 0.07010225382962612
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "83e02845524da808a5d8c256c3e5e6a8c9ec5ea227db506c087f6912d22cd0f8",
    "id": "515f37efc95bde9cd30f184b397d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 3.936227321624756,
    "wer": 0.0,
    "identity": 0.6185016632080078,
    "voiceclap": -0.10713538527488708,
    "fitness": 0.7646897960252497,
    "legacy_fitness": 0.7646897960252497,
    "target_cdf": 0.8855360207054879,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4613871574401855,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6565476192317893,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4613871574401855,
        "normalized": 0.8855360207054879,
        "contribution": 0.44276801035274393
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.34792423248291,
        "normalized": 0.7516488247877673,
        "contribution": 0.1127473237181651
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.991138219833374,
        "normalized": 0.688706627635808,
        "contribution": 0.06887066276358081
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2222a12734fde784768d96a677d4d04c09adc9211501c2c75415c915f2a993b5",
    "id": "51824d85d1533f90296168cf822d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 4.413206577301025,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.20672684907913208,
    "fitness": 0.7992236432158919,
    "legacy_fitness": 0.7992236432158919,
    "target_cdf": 0.905324255769247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.575014352798462,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6399196475464819,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.575014352798462,
        "normalized": 0.905324255769247,
        "contribution": 0.4526621278846235
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.285654067993164,
        "normalized": 0.7340144537731833,
        "contribution": 0.11010216806597749
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.3954195976257324,
        "normalized": 0.4853935573739077,
        "contribution": 0.04853935573739077
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a949cf5ea6032c217be19ea65350028d6cb3d3f6497761008e802726606a9c6f",
    "id": "51b08d0251c133b5a7247040c5f1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03899499028921127,
    "fitness": 0.8257373902023536,
    "legacy_fitness": 0.8257373902023536,
    "target_cdf": 0.9922796672226393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7275185585021973,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7306687786145152,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7275185585021973,
        "normalized": 0.9922796672226393,
        "contribution": 0.49613983361131964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5038483142852783,
        "normalized": 0.08701389328623654,
        "contribution": 0.021753473321559136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1099517345428467,
        "normalized": 0.7249653197867423,
        "contribution": 0.07249653197867423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7012ec1cd29ab73f1ec82b08cb1dbb0a360d36e685c0c03800e32d17257d2c22",
    "id": "51c68e96155f2e5cb6b983a7cfa0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.545003414154053,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.02271423302590847,
    "fitness": 0.8291652468378246,
    "legacy_fitness": 0.8291652468378246,
    "target_cdf": 0.962564179804901,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.06198787689209,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7292365168705712,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.06198787689209,
        "normalized": 0.962564179804901,
        "contribution": 0.4812820899024505
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9109374284744263,
        "normalized": 0.18656518874247485,
        "contribution": 0.04664129718561871
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.525838851928711,
        "normalized": 0.7984049796580064,
        "contribution": 0.11976074694870095
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4487907886505127,
        "normalized": 0.81552382833801,
        "contribution": 0.08155238283380101
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "79d24550df0e32546b9008a24b10f65589caa26be742f3ad60b15e648f5e3827",
    "id": "52378b4fecce55b735aa09c78934",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": 0.5022302865982056,
    "voiceclap": -0.08876892179250717,
    "fitness": 0.7439528303377995,
    "legacy_fitness": 0.7439528303377995,
    "target_cdf": 0.8081485027140713,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.1164474487304688,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5990371447680315,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.1164474487304688,
        "normalized": 0.8081485027140713,
        "contribution": 0.4040742513570357
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8702285289764404,
        "normalized": 0.17425071298327421,
        "contribution": 0.043562678245818554
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8542113304138184,
        "normalized": 0.5971951852730616,
        "contribution": 0.08957927779095924
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7755134105682373,
        "normalized": 0.6182093737421804,
        "contribution": 0.061820937374218045
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9b3e597e7cd191b6ee192d8f2a987320b31d460f9fa3a3a6e2e960ac2373df13",
    "id": "527e03ec4fe1b4f80a4dd7b8fced",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.109982967376709,
    "genuineness": 4.086852550506592,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": 0.0034197289496660233,
    "fitness": 0.754955280025468,
    "legacy_fitness": 0.754955280025468,
    "target_cdf": 0.9716911200959037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.191847801208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7491919924535275,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.191847801208496,
        "normalized": 0.9716911200959037,
        "contribution": 0.4858455600479519
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0568110942840576,
        "normalized": 0.23494557063169752,
        "contribution": 0.05873639265792438
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8193979263305664,
        "normalized": 0.8629676950479268,
        "contribution": 0.12944515425718903
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.202362537384033,
        "normalized": 0.7516488549046221,
        "contribution": 0.07516488549046221
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c13c81e250a3a14055b97a20ec7029f8abb6de9199f85ddd71b679b4415fda3a",
    "id": "527ffa65e798df0f3fc805be5167",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": 0.675062358379364,
    "voiceclap": -0.0030066687613725662,
    "fitness": 0.8260187064674552,
    "legacy_fitness": 0.8260187064674552,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7580110973202981,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9346842765808105,
        "normalized": 0.19399141727969682,
        "contribution": 0.048497854319924205
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.917250871658325,
        "normalized": 0.8809377370125135,
        "contribution": 0.13214066055187704
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4575917720794678,
        "normalized": 0.8175982255366333,
        "contribution": 0.08175982255366333
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "92861b2cb085b4751d3bd1a5fde4842c72bda0096350eb11975dbc1114b1df8e",
    "id": "528ca81e94afa7e8c67b64c9531a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 4.193545341491699,
    "wer": 0.0,
    "identity": 0.640038788318634,
    "voiceclap": -0.1525629162788391,
    "fitness": 0.826231257485774,
    "legacy_fitness": 0.826231257485774,
    "target_cdf": 0.9919413812308048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7112860679626465,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7340255890188473,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7112860679626465,
        "normalized": 0.9919413812308048,
        "contribution": 0.4959706906154024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9439380168914795,
        "normalized": 0.8855360174979171,
        "contribution": 0.13283040262468757
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.030742645263672,
        "normalized": 0.7010225382962612,
        "contribution": 0.07010225382962612
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2821673ef573a221b5ab7d94f1bd385821090c7cb0bb107f6dd9200bd5e5f78f",
    "id": "52dc8d1f4ac9dc38b5e5327a905d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.68307638168335,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.20225846767425537,
    "fitness": 0.8025848408255862,
    "legacy_fitness": 0.8025848408255862,
    "target_cdf": 0.9026628930794968,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.558781862258911,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6960167191601234,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.558781862258911,
        "normalized": 0.9026628930794968,
        "contribution": 0.4513314465397484
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0907351970672607,
        "normalized": 0.24711589066920464,
        "contribution": 0.06177897266730116
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.365715742111206,
        "normalized": 0.7565702742161262,
        "contribution": 0.11348554113241893
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.008740186691284,
        "normalized": 0.6942075882065499,
        "contribution": 0.069420758820655
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6f64ef33c9d784bd5077fa526a82b5d526e84b42c12e988ee628b6989ce4f2fc",
    "id": "52e4f070239981bd25f60ac5f70a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.193545341491699,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": -0.017192387953400612,
    "fitness": 0.7909999468871239,
    "legacy_fitness": 0.7909999468871239,
    "target_cdf": 0.9768247481628979,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.281126022338867,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7556461977938669,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.281126022338867,
        "normalized": 0.9768247481628979,
        "contribution": 0.48841237408144894
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.951646327972412,
        "normalized": 0.19940455387819955,
        "contribution": 0.04985113846954989
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.651214122772217,
        "normalized": 0.8595110255157367,
        "contribution": 0.08595110255157368
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "284b9878379af5f452350f91986e12983b158ed3c146005fc4be8a01bd9b1f94",
    "id": "530d9863659311b66bde0af24777",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.111956596374512,
    "wer": 0.0,
    "identity": 0.7788170576095581,
    "voiceclap": -0.012116985395550728,
    "fitness": 0.8008926699393741,
    "legacy_fitness": 0.8008926699393741,
    "target_cdf": 0.9409148593413338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.834733486175537,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6895646734996266,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.834733486175537,
        "normalized": 0.9409148593413338,
        "contribution": 0.4704574296706669
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6414833068847656,
        "normalized": 0.8257492948479319,
        "contribution": 0.12386239422718978
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9207301139831543,
        "normalized": 0.6662885374327974,
        "contribution": 0.06662885374327975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c2cbb91e880900b536bc24f7bf490c91b36a4a4be037b6801523b2bf8efa55f1",
    "id": "531a3c2617cce3e6d82c708c858c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.2249250411987305,
    "wer": 0.0,
    "identity": 0.5442113876342773,
    "voiceclap": -0.027925346046686172,
    "fitness": 0.7433203284182561,
    "legacy_fitness": 0.7433203284182561,
    "target_cdf": 0.812384824800176,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.1326799392700195,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5953531489746225,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.1326799392700195,
        "normalized": 0.812384824800176,
        "contribution": 0.406192412400088
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.917722225189209,
        "normalized": 0.18866878339787096,
        "contribution": 0.04716719584946774
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.7185513973236084,
        "normalized": 0.5505112423615445,
        "contribution": 0.08257668635423167
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7051053047180176,
        "normalized": 0.594168543708351,
        "contribution": 0.059416854370835104
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9fdee8f097393d10b13fbbcb11f0abf733eec32216bb1b03297662e8649c202c",
    "id": "53465a1b0d7fa04a64823b10da2a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.570657730102539,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.033341728150844574,
    "fitness": 0.8156695922229749,
    "legacy_fitness": 0.8156695922229749,
    "target_cdf": 0.9585742155738795,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0132906436920166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7096669895226143,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0132906436920166,
        "normalized": 0.9585742155738795,
        "contribution": 0.4792871077869397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7616714239120483,
        "normalized": 0.1440043780288135,
        "contribution": 0.036001094507203374
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.71264910697937,
        "normalized": 0.8413447460685429,
        "contribution": 0.12620171191028143
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9691355228424072,
        "normalized": 0.6817707531818977,
        "contribution": 0.06817707531818977
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7276c10f687ebf0b20e84a13119832e0baf80628aa4ae7f82f97a63fd7d46001",
    "id": "53537119ee42a5ec19d5d19a3ebf",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.260893821716309,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.029577042907476425,
    "fitness": 0.8205669182663009,
    "legacy_fitness": 0.8205669182663009,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8035146489885527,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1687605381011963,
        "normalized": 0.2763396656306163,
        "contribution": 0.06908491640765407
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.477681636810303,
        "normalized": 0.9527255009885056,
        "contribution": 0.14290882514827583
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.22327995300293,
        "normalized": 0.9436232660045483,
        "contribution": 0.09436232660045484
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a30dd0c514c066362b4f0213a9527900f674f24854c9522d5cea1f16ec1ec56c",
    "id": "535f9463f5fb9775254d60bd67a5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.09409761428833,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": 0.5979136228561401,
    "voiceclap": 0.058065664023160934,
    "fitness": 0.8052378450538056,
    "legacy_fitness": 0.8052378450538056,
    "target_cdf": 0.9929189433068706,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.759983539581299,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7493306599834674,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.759983539581299,
        "normalized": 0.9929189433068706,
        "contribution": 0.4964594716534353
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5920509099960327,
        "normalized": 0.10422977602228944,
        "contribution": 0.02605744400557236
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.406516075134277,
        "normalized": 0.9462328279462877,
        "contribution": 0.14193492419194315
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5984079837799072,
        "normalized": 0.8487882013251657,
        "contribution": 0.08487882013251657
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6789e0fe819cc0e46a76d852b799166556ec2067f692a34fda0a94dcf9aac71b",
    "id": "53621c14de31d1e4b226e66c0a07",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.507347106933594,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10139834880828857,
    "fitness": 0.7785855202478427,
    "legacy_fitness": 0.7785855202478427,
    "target_cdf": 0.8697054919803655,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3802249431610107,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.721048280480517,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3802249431610107,
        "normalized": 0.8697054919803655,
        "contribution": 0.43485274599018275
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.615286350250244,
        "normalized": 0.46828170559335164,
        "contribution": 0.11707042639833791
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.914257526397705,
        "normalized": 0.6174644295245129,
        "contribution": 0.09261966442867693
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.250767946243286,
        "normalized": 0.765054436633194,
        "contribution": 0.0765054436633194
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "51c8d75f1e3fb911272911051b2d88e8da8c6b6deedebf774d9829217d29ff7f",
    "id": "536d20873079e17e7cac751779ee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.362998008728027,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12964247167110443,
    "fitness": 0.8190451622955071,
    "legacy_fitness": 0.8190451622955071,
    "target_cdf": 0.9527255219125881,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9483609199523926,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.70571233155928,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9483609199523926,
        "normalized": 0.9527255219125881,
        "contribution": 0.47636276095629404
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7215447425842285,
        "normalized": 0.8432277384318043,
        "contribution": 0.12648416076477065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9691355228424072,
        "normalized": 0.6817707531818977,
        "contribution": 0.06817707531818977
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ad588f465799aed871ceba61cf4ebd42d55aed53212dcd648ae331f52d8ce51e",
    "id": "538aeb69c43dca9b5e2e924cf090",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": 0.5647817850112915,
    "voiceclap": 0.06529650092124939,
    "fitness": 0.796928753596261,
    "legacy_fitness": 0.796928753596261,
    "target_cdf": 0.9706627967789073,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1756153106689453,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7607005012808359,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1756153106689453,
        "normalized": 0.9706627967789073,
        "contribution": 0.48533139838945366
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.975393295288086,
        "normalized": 0.20713393848934814,
        "contribution": 0.051783484622337034
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.210809707641602,
        "normalized": 0.9247119875768577,
        "contribution": 0.13870679813652864
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5984079837799072,
        "normalized": 0.8487882013251657,
        "contribution": 0.08487882013251657
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a911b7b0f36d07a93ae1f5f40dce18c77e4eba89ffbb1f948c4742941bcc47a3",
    "id": "53e8ff74cff7b3a41b41a699e54d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.149612903594971,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0959780290722847,
    "fitness": 0.8333226463144099,
    "legacy_fitness": 0.8333226463144099,
    "target_cdf": 0.999062582044815,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.441746711730957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7848055956947112,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.441746711730957,
        "normalized": 0.999062582044815,
        "contribution": 0.4995312910224075
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9143297672271729,
        "normalized": 0.18761514602105922,
        "contribution": 0.046903786505264805
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.851302146911621,
        "normalized": 0.9772498567468411,
        "contribution": 0.14658747851202616
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.003254413604736,
        "normalized": 0.9178303965501271,
        "contribution": 0.09178303965501272
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7eb1d2b19bdabba274eb57b6f23b7e48f95ea8a0ae60842eecc40915ebc90d8b",
    "id": "546810f085de20b39d3896ad1cd5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.055472373962402,
    "wer": 0.0,
    "identity": 0.6600226163864136,
    "voiceclap": -0.10290581732988358,
    "fitness": 0.7331296757737886,
    "legacy_fitness": 0.7331296757737886,
    "target_cdf": 0.8049331497907819,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.1042730808258057,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6037489948037186,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.1042730808258057,
        "normalized": 0.8049331497907819,
        "contribution": 0.40246657489539095
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.192249059677124,
        "normalized": 0.7064243659623586,
        "contribution": 0.10596365489435379
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6236958503723145,
        "normalized": 0.5659276669838458,
        "contribution": 0.05659276669838458
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "30debaa5900cc141a5e8616a0802e294154a86f1808468d56e42cb1b76bbf6e7",
    "id": "5490b60c86cbf2f5fa4aa2d67fee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.173524379730225,
    "genuineness": 3.7793264389038086,
    "wer": 0.0,
    "identity": 0.5121695399284363,
    "voiceclap": 0.03927452117204666,
    "fitness": 0.81145127352509,
    "legacy_fitness": 0.81145127352509,
    "target_cdf": 0.9992824581550183,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.522909164428711,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8428307551546822,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.522909164428711,
        "normalized": 0.9992824581550183,
        "contribution": 0.49964122907750913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4248874187469482,
        "normalized": 0.38253558321146375,
        "contribution": 0.09563389580286594
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.438420295715332,
        "normalized": 0.9940589205534311,
        "contribution": 0.14910883808301467
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.865753650665283,
        "normalized": 0.9844679219129244,
        "contribution": 0.09844679219129245
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9b79f5e1f0b370e9f00d2302065373653a32bc4569200038fbf18cee80e839b9",
    "id": "54ce45146c3b0b8f34eecefc5bf7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 3.565941095352173,
    "wer": 0.0,
    "identity": 0.6135305166244507,
    "voiceclap": 0.10744301974773407,
    "fitness": 0.803584518925479,
    "legacy_fitness": 0.803584518925479,
    "target_cdf": 0.9965654549919187,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.019702911376953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7809482611703388,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.019702911376953,
        "normalized": 0.9965654549919187,
        "contribution": 0.49828272749595937
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.887190580368042,
        "normalized": 0.17931761586218503,
        "contribution": 0.04482940396554626
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.94025993347168,
        "normalized": 0.9811510867437907,
        "contribution": 0.1471726630115686
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9240455627441406,
        "normalized": 0.9066346669726454,
        "contribution": 0.09066346669726455
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1acb8f478f0aff085f083e24ef0591e9a0a3b3278829d44b113e83db961061d2",
    "id": "54eddadad76bc0e8ff8ec5c15bae",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": 0.429105669260025,
    "voiceclap": -0.1688079535961151,
    "fitness": 0.8338153410445904,
    "legacy_fitness": 0.8338153410445904,
    "target_cdf": 0.9932207044941694,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7762160301208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7359626888176978,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7762160301208496,
        "normalized": 0.9932207044941694,
        "contribution": 0.4966103522470847
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.193018436431885,
        "normalized": 0.9224687261883889,
        "contribution": 0.13837030892825833
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.105551242828369,
        "normalized": 0.7236603178386457,
        "contribution": 0.07236603178386457
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9ab869637ab9531355c6f9e574f09f24125111380a349b040b04d4ab41f2b69b",
    "id": "55629c05dbc5dadf1fa6bbe9e4da",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 4.218649387359619,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.21573692560195923,
    "fitness": 0.7827201939102313,
    "legacy_fitness": 0.7827201939102313,
    "target_cdf": 0.8885304563197098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4776196479797363,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6465889686282956,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4776196479797363,
        "normalized": 0.8885304563197098,
        "contribution": 0.4442652281598549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7209625244140625,
        "normalized": 0.13363422338747727,
        "contribution": 0.03340855584686932
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2945499420166016,
        "normalized": 0.736571978107565,
        "contribution": 0.11048579671613475
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.676501989364624,
        "normalized": 0.5842938790543667,
        "contribution": 0.05842938790543667
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e51c37d5b1f07911cc75f42a2bdb8b00551d3128ee82dd1f8f405d885dbec4b9",
    "id": "55650ef40ebf59b58e336708fc17",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": 0.5836091041564941,
    "voiceclap": -0.10932472348213196,
    "fitness": 0.8232686382670478,
    "legacy_fitness": 0.8232686382670478,
    "target_cdf": 0.9716911200959037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.191847801208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6825262407525312,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.191847801208496,
        "normalized": 0.9716911200959037,
        "contribution": 0.4858455600479519
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6802536249160767,
        "normalized": 0.12378946157536314,
        "contribution": 0.030947365393840784
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2322797775268555,
        "normalized": 0.7184101418133952,
        "contribution": 0.10776152127200929
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6633005142211914,
        "normalized": 0.5797179403872929,
        "contribution": 0.05797179403872929
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "be03352ab160965ae45043712ade775d3ffa4a8bc0f62fd900debd16fcf09e82",
    "id": "55c3de142e2bc1e92adfe257478f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.538887023925781,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.15814203023910522,
    "fitness": 0.8073235837265742,
    "legacy_fitness": 0.8073235837265742,
    "target_cdf": 0.9235966379731679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6967577934265137,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6529929192606253,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6967577934265137,
        "normalized": 0.9235966379731679,
        "contribution": 0.46179831898658397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1032917499542236,
        "normalized": 0.678978341920392,
        "contribution": 0.1018467512880588
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5807909965515137,
        "normalized": 0.5508977139443143,
        "contribution": 0.055089771394431435
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f82fe6bdb8c35b2a48ce97acafd0cc74532b1aedced65af8593abfaec999fae1",
    "id": "56089d5cbba91b474508298de134",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": 0.6937824487686157,
    "voiceclap": -0.009839002043008804,
    "fitness": 0.7783775630846961,
    "legacy_fitness": 0.7783775630846961,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7481045344848024,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.268836498260498,
        "normalized": 0.3161415788970373,
        "contribution": 0.07903539472425933
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.632587432861328,
        "normalized": 0.8237335195645583,
        "contribution": 0.12356002793468374
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6248111724853516,
        "normalized": 0.8542158963153907,
        "contribution": 0.08542158963153908
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0087d69ee394ef938eda16f346635d3fe9b5bab29cb3f4c30317db0cd7783f0a",
    "id": "56b46fad03d3511748d2fdfbf760",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.923675537109375,
    "wer": 0.0,
    "identity": 0.6505627632141113,
    "voiceclap": 0.016144676133990288,
    "fitness": 0.8145906800269876,
    "legacy_fitness": 0.8145906800269876,
    "target_cdf": 0.9832067001342365,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.419102191925049,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7511209295042799,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.419102191925049,
        "normalized": 0.9832067001342365,
        "contribution": 0.49160335006711825
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.707392930984497,
        "normalized": 0.130294529860791,
        "contribution": 0.03257363246519775
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.442098617553711,
        "normalized": 0.9495623364922869,
        "contribution": 0.14243435047384304
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.580805778503418,
        "normalized": 0.8450959649812083,
        "contribution": 0.08450959649812084
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c744f2526b5bf3bbfaa5f72ac632e48fdfe81f519eb9e6bbc8f82e5167d22a06",
    "id": "56c3d0641ae048937301e71a001a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.898571252822876,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.030300062149763107,
    "fitness": 0.7880546959870565,
    "legacy_fitness": 0.7880546959870565,
    "target_cdf": 0.929049708969098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7373390197753906,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7287804827557193,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7373390197753906,
        "normalized": 0.929049708969098,
        "contribution": 0.464524854484549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.039849042892456,
        "normalized": 0.2289868888511522,
        "contribution": 0.05724672221278805
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6236917972564697,
        "normalized": 0.8217031044875092,
        "contribution": 0.12325546567312637
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5456018447875977,
        "normalized": 0.8375344038525582,
        "contribution": 0.08375344038525583
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "de3cc1437c1bd4e2bbc0c80e9b5e15a3c48fd6f56c110ee0ea62a6d45ae30d5f",
    "id": "572a98a8968dd5b1cefd0d1d8c01",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 3.8546390533447266,
    "wer": 0.0,
    "identity": 0.5623583197593689,
    "voiceclap": 0.11118116974830627,
    "fitness": 0.8192750089135765,
    "legacy_fitness": 0.8192750089135765,
    "target_cdf": 0.9999169335299728,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.123509883880615,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7917282053798,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.123509883880615,
        "normalized": 0.9999169335299728,
        "contribution": 0.4999584667649864
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9448615312576294,
        "normalized": 0.19722848067005766,
        "contribution": 0.049307120167514415
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.1003828048706055,
        "normalized": 0.986748131774742,
        "contribution": 0.1480122197662113
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.232080936431885,
        "normalized": 0.9445039868108795,
        "contribution": 0.09445039868108795
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cbdfa301aa44f9efa83efe86b6836037d1d1a75e3f19695343f2e395f690acd9",
    "id": "57342a43612799b5fa45c69aaf92",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.117925643920898,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.128897562623024,
    "fitness": 0.8225696558460928,
    "legacy_fitness": 0.8225696558460928,
    "target_cdf": 0.9929189433068706,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.759983539581299,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8521769648965994,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.759983539581299,
        "normalized": 0.9929189433068706,
        "contribution": 0.4964594716534353
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.584754705429077,
        "normalized": 0.4543231757231024,
        "contribution": 0.1135807939307756
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.869093894958496,
        "normalized": 0.9780804006212112,
        "contribution": 0.14671206009318166
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.337693214416504,
        "normalized": 0.9542463921920685,
        "contribution": 0.09542463921920685
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a304ebe7f872c7aadd25a99a5d42168f0bb4b6db79d199d8d272d9167aefa943",
    "id": "577b20e1c04d27df393a39152a55",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 3.9425034523010254,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07968109846115112,
    "fitness": 0.8088600146307275,
    "legacy_fitness": 0.8088600146307275,
    "target_cdf": 0.9638222590155852,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0782203674316406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8153063487218953,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0782203674316406,
        "normalized": 0.9638222590155852,
        "contribution": 0.4819111295077926
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5419256687164307,
        "normalized": 0.4348410174256361,
        "contribution": 0.10871025435640902
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.997312545776367,
        "normalized": 0.8943502454552662,
        "contribution": 0.13415253681828992
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9152445793151855,
        "normalized": 0.9053242803940382,
        "contribution": 0.09053242803940383
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "96be36d4ebc3abfba8045638ed85d71d6e71dd1e3787a8496f3b830c28dde2c0",
    "id": "57eaf84db88f6602284260a13739",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": 0.5321549773216248,
    "voiceclap": -0.021186642348766327,
    "fitness": 0.8294377888739868,
    "legacy_fitness": 0.8294377888739868,
    "target_cdf": 0.9772498618564534,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2892422676086426,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7550278764710912,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2892422676086426,
        "normalized": 0.9772498618564534,
        "contribution": 0.4886249309282267
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9855704307556152,
        "normalized": 0.2105000137670302,
        "contribution": 0.05262500344175755
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9884166717529297,
        "normalized": 0.8929163005060587,
        "contribution": 0.1339374450759088
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.378382682800293,
        "normalized": 0.7984049702519815,
        "contribution": 0.07984049702519816
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "88c1fa45a8658e7b86e6464c6f98679b4bd8c698e36e281376dc78b5c59b913a",
    "id": "57f4effed7155047e60757af0e1d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": 0.5942701101303101,
    "voiceclap": 0.07679415494203568,
    "fitness": 0.8398366658847838,
    "legacy_fitness": 0.8398366658847838,
    "target_cdf": 0.9965654549919187,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.019702911376953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7876867170752258,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.019702911376953,
        "normalized": 0.9965654549919187,
        "contribution": 0.49828272749595937
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1721529960632324,
        "normalized": 0.2776477208574187,
        "contribution": 0.06941193021435467
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.466392755508423,
        "normalized": 0.8196579878328937,
        "contribution": 0.08196579878328937
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1acbe7f861b1c11a9caed3a3473b1eee587d7e5031105bc8f2b59fcfab7f63a3",
    "id": "585fd72ab16ec303bc758f227b15",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": 0.5656378269195557,
    "voiceclap": 0.04967792332172394,
    "fitness": 0.8280408856774341,
    "legacy_fitness": 0.8280408856774341,
    "target_cdf": 0.9994538858542781,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.604071140289307,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7959071052385982,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.604071140289307,
        "normalized": 0.9994538858542781,
        "contribution": 0.49972694292713904
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.033064126968384,
        "normalized": 0.2266273368759334,
        "contribution": 0.05665683421898335
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.886885643005371,
        "normalized": 0.97888519494114,
        "contribution": 0.146832779241171
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.073662757873535,
        "normalized": 0.926905488513048,
        "contribution": 0.0926905488513048
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1c29eda385281694d0e43022d3703260515ebd2e2a5e4da03af70f7ef61d4ecc",
    "id": "58712af1e16ae6833d12841a6b7d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": 0.606049656867981,
    "voiceclap": -0.02860785648226738,
    "fitness": 0.8183508907185905,
    "legacy_fitness": 0.8183508907185905,
    "target_cdf": 0.946232810502631,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8834309577941895,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7692063516819047,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8834309577941895,
        "normalized": 0.946232810502631,
        "contribution": 0.4731164052513155
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3960518836975098,
        "normalized": 0.36993279497136494,
        "contribution": 0.09248319874284124
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.508047342300415,
        "normalized": 0.7939810079308474,
        "contribution": 0.1190971511896271
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.580805778503418,
        "normalized": 0.8450959649812083,
        "contribution": 0.08450959649812084
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7aace50e4e99620b3f5eeba7d6858d4b6b8c9f889ab6d6b154d6463a2e1b97c9",
    "id": "587cc162c8231733b8b51e741484",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.12167966365814209,
    "fitness": 0.8174748899021343,
    "legacy_fitness": 0.8174748899021343,
    "target_cdf": 0.9915901345965457,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6950535774230957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7525785817732423,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6950535774230957,
        "normalized": 0.9915901345965457,
        "contribution": 0.49579506729827283
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.5132646560668945,
        "normalized": 0.9557276577709561,
        "contribution": 0.14335914866564342
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5059971809387207,
        "normalized": 0.8287452780397044,
        "contribution": 0.08287452780397045
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3a64c778b5f7f0393c7cc43da358ba971dfbfe3869ff49e56463253cac0c0aca",
    "id": "5930cddfa791d81dfc98bec9664e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": 0.7428895235061646,
    "voiceclap": -0.03196481615304947,
    "fitness": 0.7597870688171604,
    "legacy_fitness": 0.7597870688171604,
    "target_cdf": 0.836572595542951,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.230074644088745,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6088479856208728,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.230074644088745,
        "normalized": 0.836572595542951,
        "contribution": 0.4182862977714755
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.693823218345642,
        "normalized": 0.127012998991818,
        "contribution": 0.0317532497479545
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0299019813537598,
        "normalized": 0.6555656036609419,
        "contribution": 0.09833484054914128
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7359087467193604,
        "normalized": 0.6047359755230154,
        "contribution": 0.06047359755230154
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bc42c6d94aa75e06eed3f18dcb67b2fb76f28242624f1f361b6265b500a2af7b",
    "id": "594b13a89024c57cc888e79c4b32",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.406930446624756,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12237809598445892,
    "fitness": 0.8101075121161386,
    "legacy_fitness": 0.8101075121161386,
    "target_cdf": 0.9371461924705993,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8022687435150146,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7087012689438292,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8022687435150146,
        "normalized": 0.9371461924705993,
        "contribution": 0.46857309623529964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0194945335388184,
        "normalized": 0.22194973203149376,
        "contribution": 0.05548743300787344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.321237087249756,
        "normalized": 0.744168371280294,
        "contribution": 0.1116252556920441
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.127553939819336,
        "normalized": 0.7301548400861202,
        "contribution": 0.07301548400861202
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "41c04d96291f4d775c94fa08f60a0ca134d328a43139a6ec88367fc6552648d3",
    "id": "599d343b70c622ef531d76b3b21d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.557555198669434,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11681827902793884,
    "fitness": 0.841519810001325,
    "legacy_fitness": 0.841519810001325,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7296500526350592,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7037534713745117,
        "normalized": 0.8394469854437661,
        "contribution": 0.1259170478165649
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0351431369781494,
        "normalized": 0.7023772084492015,
        "contribution": 0.07023772084492015
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "908072f6a06d2d6d8a1ee3ce79e9d969166fd33a2d87564ad8a22e5a034a563b",
    "id": "59a93f51defcb855c892a615b13c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.432034492492676,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1830248087644577,
    "fitness": 0.8130378474626692,
    "legacy_fitness": 0.8130378474626692,
    "target_cdf": 0.9409148593413338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.834733486175537,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7209692703400191,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.834733486175537,
        "normalized": 0.9409148593413338,
        "contribution": 0.4704574296706669
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.066988229751587,
        "normalized": 0.2385614708758641,
        "contribution": 0.059640367718966025
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3390283584594727,
        "normalized": 0.7491683218823387,
        "contribution": 0.1123752482823508
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3255765438079834,
        "normalized": 0.7849622466803536,
        "contribution": 0.07849622466803537
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "99d79b0a9a0e12ba22bf881162ed080a8a5aee276f0d3f3f9555c8685d24e2c9",
    "id": "5ab18a4adcd6a40fc8ea02e34d0b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": 0.5491657853126526,
    "voiceclap": -0.06718175113201141,
    "fitness": 0.7880183396146638,
    "legacy_fitness": 0.7880183396146638,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6856569387941595,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.71264910697937,
        "normalized": 0.8413447460685429,
        "contribution": 0.12620171191028143
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9779367446899414,
        "normalized": 0.6845529821840911,
        "contribution": 0.06845529821840911
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "068bcfe74fa656ab4a5e46ef6a407c6fbd6ccab217c879e8293dab5d47719a9c",
    "id": "5adcd5cca88640066e6e2d83b233",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": 0.6941993832588196,
    "voiceclap": 0.03504546731710434,
    "fitness": 0.7746475794241676,
    "legacy_fitness": 0.7746475794241676,
    "target_cdf": 0.880937722986112,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4370384216308594,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6937554115522561,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4370384216308594,
        "normalized": 0.880937722986112,
        "contribution": 0.440468861493056
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1450138092041016,
        "normalized": 0.26726902011026354,
        "contribution": 0.06681725502756589
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3078935146331787,
        "normalized": 0.7403845185212657,
        "contribution": 0.11105767777818985
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2111635208129883,
        "normalized": 0.754116172534443,
        "contribution": 0.07541161725344431
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "11f512554aa825eb475387cc1e77e9dd3de175dd5f9e2533e48a3d56bf1044b2",
    "id": "5b6736c0032b24c08fa629b578b9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.024253398180007935,
    "fitness": 0.8034071546340417,
    "legacy_fitness": 0.8034071546340417,
    "target_cdf": 0.9662346008508267,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.110685348510742,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.741095415341717,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.110685348510742,
        "normalized": 0.9662346008508267,
        "contribution": 0.48311730042541334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.917722225189209,
        "normalized": 0.18866878339787096,
        "contribution": 0.04716719584946774
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8816680908203125,
        "normalized": 0.8746060247794507,
        "contribution": 0.1311909037169176
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.369581699371338,
        "normalized": 0.7962001534991823,
        "contribution": 0.07962001534991824
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a4251aeb3b49a621fd02cd1d840ffb768af1ba5e511e8bf66b9544890964d07b",
    "id": "5bb6a58d50827a62ab35faabde31",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.469690799713135,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.2459096610546112,
    "fitness": 0.8002453388808148,
    "legacy_fitness": 0.8002453388808148,
    "target_cdf": 0.912986084230339,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.623711585998535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6545573816807757,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.623711585998535,
        "normalized": 0.912986084230339,
        "contribution": 0.4564930421151695
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1099634170532227,
        "normalized": 0.6810736024334292,
        "contribution": 0.10216104036501437
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8327198028564453,
        "normalized": 0.6374167680329237,
        "contribution": 0.06374167680329237
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9decca3495d2e57d70be4828fe7955841172349bff967bd42bb16413ff3a7361",
    "id": "5bed46de4d3c60823f65ee56e19e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.400654315948486,
    "wer": 0.0,
    "identity": 0.41842520236968994,
    "voiceclap": 0.09722457081079483,
    "fitness": 0.8405170056862941,
    "legacy_fitness": 0.8405170056862941,
    "target_cdf": 0.9992824581550183,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.522909164428711,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7583143979150158,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.522909164428711,
        "normalized": 0.9992824581550183,
        "contribution": 0.49964122907750913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.637804985046387,
        "normalized": 0.965045519092528,
        "contribution": 0.1447568278638792
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5279998779296875,
        "normalized": 0.8336650296827179,
        "contribution": 0.0833665029682718
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f3ae75765ac215c7f9887e3ae6c0587aa241ea8748b9435614c370fd06b5eb2e",
    "id": "5c326936abd40e23d97b0b0a05b6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": 0.7517085075378418,
    "voiceclap": 0.038323406130075455,
    "fitness": 0.8304506517883898,
    "legacy_fitness": 0.8304506517883898,
    "target_cdf": 0.9981710548076552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.230724811553955,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7900341144584342,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.230724811553955,
        "normalized": 0.9981710548076552,
        "contribution": 0.4990855274038276
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0568110942840576,
        "normalized": 0.23494557063169752,
        "contribution": 0.05873639265792438
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.5132646560668945,
        "normalized": 0.9557276577709561,
        "contribution": 0.14335914866564342
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8096323013305664,
        "normalized": 0.8885304573103883,
        "contribution": 0.08885304573103883
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "aee132e50626156bf120d6b57e818fc4ed338b7c03477093ede542cb2a5323d5",
    "id": "5c45096a6d7a2346b47824709e70",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.11401297152042389,
    "fitness": 0.8111328037411777,
    "legacy_fitness": 0.8111328037411777,
    "target_cdf": 0.9935110473296658,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7924485206604004,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7813469441943153,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7924485206604004,
        "normalized": 0.9935110473296658,
        "contribution": 0.4967555236648329
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9312918186187744,
        "normalized": 0.19291961867711876,
        "contribution": 0.04822990466927969
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.753449440002441,
        "normalized": 0.9721939188053255,
        "contribution": 0.14582908782079881
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9152445793151855,
        "normalized": 0.9053242803940382,
        "contribution": 0.09053242803940383
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "87c37878b86602a5e6b9a57d87ab5bbce00f2e8a6f01d957d271b443a3537fe4",
    "id": "5cda7ad63b0954640bbfdfd35a87",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.218649387359619,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.0432136207818985,
    "fitness": 0.7450143093247535,
    "legacy_fitness": 0.7450143093247535,
    "target_cdf": 0.9247120186784414,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.704874038696289,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6567393744942688,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.704874038696289,
        "normalized": 0.9247120186784414,
        "contribution": 0.4623560093392207
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.693823218345642,
        "normalized": 0.127012998991818,
        "contribution": 0.0317532497479545
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3123412132263184,
        "normalized": 0.7416489601774914,
        "contribution": 0.1112473440266237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.475728750228882,
        "normalized": 0.5138277138046985,
        "contribution": 0.05138277138046985
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ce595b97b3964a14ade1404bfea2199e40fa776b98b89ff9ca0490faf89ae262",
    "id": "5d0f42a12cd89e6b61cef0269224",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.482243061065674,
    "wer": 0.06666666666666667,
    "identity": 0.6034767627716064,
    "voiceclap": -0.06581717729568481,
    "fitness": 0.7284834872380337,
    "legacy_fitness": 0.7284834872380337,
    "target_cdf": 0.8697054919803655,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3802249431610107,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6654897390428364,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3802249431610107,
        "normalized": 0.8697054919803655,
        "contribution": 0.43485274599018275
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1755452156066895,
        "normalized": 0.27895870121176836,
        "contribution": 0.06973967530294209
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0899481773376465,
        "normalized": 0.6747706587485832,
        "contribution": 0.10121559881228748
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.712806224822998,
        "normalized": 0.59681718937424,
        "contribution": 0.059681718937424
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d386f111d3dac7ce034812b0629bebcec7b22f852d8e8e04da36278ab2c2008d",
    "id": "5d3febd9ee450a445f0a5276eed1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11570194363594055,
    "fitness": 0.8195720206366246,
    "legacy_fitness": 0.8195720206366246,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7802688298813748,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1280517578125,
        "normalized": 0.2608832234596053,
        "contribution": 0.06522080586490132
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8448362350463867,
        "normalized": 0.8943502118372078,
        "contribution": 0.08943502118372078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2765031370dddc0abfc70812cf8c8ace897e59ff8aabccb7dbf836ab231b4a3c",
    "id": "5d5bde75925543455c696f62534c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.162164688110352,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.049065522849559784,
    "fitness": 0.8092665284888942,
    "legacy_fitness": 0.8092665284888942,
    "target_cdf": 0.947918719056646,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8996634483337402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7126737676175089,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8996634483337402,
        "normalized": 0.947918719056646,
        "contribution": 0.473959359528323
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8363044261932373,
        "normalized": 0.16439289419288416,
        "contribution": 0.04109822354822104
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6147961616516113,
        "normalized": 0.8196580135026951,
        "contribution": 0.12294870202540426
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.184760332107544,
        "normalized": 0.7466748251556059,
        "contribution": 0.07466748251556059
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "18cd5ef76a180943ad69febc4d32914df12fa944b9b8a33166d3d98df6832e86",
    "id": "5d94d7821fb104636fbb1abbdae6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": 0.674980640411377,
    "voiceclap": 0.001098395325243473,
    "fitness": 0.8265864544795465,
    "legacy_fitness": 0.8265864544795465,
    "target_cdf": 0.9995376693579576,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.652768611907959,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7650801092634608,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.652768611907959,
        "normalized": 0.9995376693579576,
        "contribution": 0.4997688346789788
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.693823218345642,
        "normalized": 0.127012998991818,
        "contribution": 0.0317532497479545
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.851302146911621,
        "normalized": 0.9772498567468411,
        "contribution": 0.14658747851202616
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7040200233459473,
        "normalized": 0.8697054632450133,
        "contribution": 0.08697054632450134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8e00c04efa8b0ff37d082b60736e2aec18a5a83967a0758ad83c9ead1b040566",
    "id": "5e6727ab4d1341933bddef2048c6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.5786004066467285,
    "genuineness": 3.9676074981689453,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07491813600063324,
    "fitness": 0.8188221862580641,
    "legacy_fitness": 0.8188221862580641,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7594479666421359,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9991400241851807,
        "normalized": 0.2150377207338816,
        "contribution": 0.0537594301834704
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.032895088195801,
        "normalized": 0.8999470264638179,
        "contribution": 0.13499205396957267
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4531912803649902,
        "normalized": 0.8165628548756955,
        "contribution": 0.08165628548756955
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "838d5e9008a67cd2adbf8ff55bbddea7dac3823b5d454c4635040be223905136",
    "id": "5eb29b42f0da78a42d31a8dd4e8e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.538887023925781,
    "genuineness": 4.268857479095459,
    "wer": 0.0,
    "identity": 0.5364359617233276,
    "voiceclap": -0.07889751344919205,
    "fitness": 0.837643932864932,
    "legacy_fitness": 0.837643932864932,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7402488410051236,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8295196294784546,
        "normalized": 0.16246558721860938,
        "contribution": 0.040616396804652344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8638763427734375,
        "normalized": 0.8713534713164917,
        "contribution": 0.13070302069747375
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.114352226257324,
        "normalized": 0.7262672786518737,
        "contribution": 0.07262672786518737
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "217e8c96eebcc521d0235e35f17618074647b20d701cd76eee71cea9613b6a15",
    "id": "5f2d84563cb18eaa0ea244e9ac21",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.229123115539551,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.08737954497337341,
    "fitness": 0.8262178169129359,
    "legacy_fitness": 0.8262178169129359,
    "target_cdf": 0.9726892523841197,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2080800533294678,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7348524394974176,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2080800533294678,
        "normalized": 0.9726892523841197,
        "contribution": 0.48634462619205987
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8363044261932373,
        "normalized": 0.16439289419288416,
        "contribution": 0.04109822354822104
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2859721183776855,
        "normalized": 0.7745472612739859,
        "contribution": 0.0774547261273986
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a02c8b7192aa80d22546a98b8177d5197cd98bac9320f87b1f921286c396f13e",
    "id": "5f783bce7ecc51dd1df792912ef0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": 0.6559209227561951,
    "voiceclap": 0.0517394095659256,
    "fitness": 0.7855238890984937,
    "legacy_fitness": 0.7855238890984937,
    "target_cdf": 0.9418287707049221,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8428497314453125,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7428421832553431,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8428497314453125,
        "normalized": 0.9418287707049221,
        "contribution": 0.47091438535246105
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0568110942840576,
        "normalized": 0.23494557063169752,
        "contribution": 0.05873639265792438
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4575917720794678,
        "normalized": 0.8175982255366333,
        "contribution": 0.08175982255366333
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "608437a5c54928e0225532ea3857091a24146a895559fc7ce666c0ab1deb0dfe",
    "id": "5fa8279e41e6dd21b28364b70688",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.7458367347717285,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.03250148147344589,
    "fitness": 0.7455294342227463,
    "legacy_fitness": 0.7455294342227463,
    "target_cdf": 0.7861019872402517,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.035285234451294,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5488206728147899,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.035285234451294,
        "normalized": 0.7861019872402517,
        "contribution": 0.39305099362012585
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.6551692485809326,
        "normalized": 0.5284161134129087,
        "contribution": 0.0792624170119363
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 1.970220685005188,
        "normalized": 0.33941416124907164,
        "contribution": 0.03394141612490716
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "529e258275a1f5f8cf5aa2286949ecf29a429a4247b0f2d764601921d3ed0c3b",
    "id": "5fbd550a95144e3f9b99112a5e7d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.017816066741943,
    "wer": 0.0,
    "identity": 0.426835298538208,
    "voiceclap": -0.09516935050487518,
    "fitness": 0.8263909019743196,
    "legacy_fitness": 0.8263909019743196,
    "target_cdf": 0.9977706984112247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.165794849395752,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7475605693878761,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.165794849395752,
        "normalized": 0.9977706984112247,
        "contribution": 0.49888534920561234
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6259750127792358,
        "normalized": 0.11146953771947371,
        "contribution": 0.027867384429868428
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.3709330558776855,
        "normalized": 0.9427315543087618,
        "contribution": 0.14140973314631425
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.360780715942383,
        "normalized": 0.7939810260608104,
        "contribution": 0.07939810260608104
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "88e8eef59980729b3684dc7e25e5a58a441bb48257de884508d0d626967294ad",
    "id": "5fdd69b015d7e4ce849bb25ef8fe",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": 0.4843508005142212,
    "voiceclap": -0.05397970229387283,
    "fitness": 0.8237616625015028,
    "legacy_fitness": 0.8237616625015028,
    "target_cdf": 0.9990117801396228,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.425514221191406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.784123241021824,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.425514221191406,
        "normalized": 0.9990117801396228,
        "contribution": 0.4995058900698114
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8973677158355713,
        "normalized": 0.18240174630185035,
        "contribution": 0.04560043657546259
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.8157196044921875,
        "normalized": 0.9755091035016348,
        "contribution": 0.14632636552524522
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.073662757873535,
        "normalized": 0.926905488513048,
        "contribution": 0.0926905488513048
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e78aac67bb5d7d46317dd3549cd555d7e59c1b7ed9366c2415a2a249f8442a95",
    "id": "5ff3433809e2310179073c6f0d3d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.55477237701416,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0371868796646595,
    "fitness": 0.8272852165646617,
    "legacy_fitness": 0.8272852165646617,
    "target_cdf": 0.9679556654189556,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1350340843200684,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7270827016565962,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1350340843200684,
        "normalized": 0.9679556654189556,
        "contribution": 0.4839778327094778
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9448615312576294,
        "normalized": 0.19722848067005766,
        "contribution": 0.049307120167514415
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.632587432861328,
        "normalized": 0.8237335195645583,
        "contribution": 0.12356002793468374
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0351431369781494,
        "normalized": 0.7023772084492015,
        "contribution": 0.07023772084492015
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fcf12e56d7692ad9f563bd28618c356ef98bb9b9b7e0b5b9208ee53627c6d07e",
    "id": "603ed14dd3da8d6bf0fbf981086e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.632867813110352,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13014811277389526,
    "fitness": 0.7767325494734054,
    "legacy_fitness": 0.7767325494734054,
    "target_cdf": 0.8487882165208668,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.282830238342285,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6004525291802234,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.282830238342285,
        "normalized": 0.8487882165208668,
        "contribution": 0.4243941082604334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.974303722381592,
        "normalized": 0.637416836817112,
        "contribution": 0.0956125255225668
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.328861951828003,
        "normalized": 0.46187817805672027,
        "contribution": 0.04618781780567203
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1b080fc298aeea2ec4e526efd4c0acf6ad0378f130661ead6c557fd067f7bf31",
    "id": "605ea94b41c30ed51ddb1ded4cd5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": 0.7948168516159058,
    "voiceclap": 0.04110018536448479,
    "fitness": 0.8167882669715574,
    "legacy_fitness": 0.8167882669715574,
    "target_cdf": 0.9998935200270991,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.058579921722412,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7743649757991453,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.058579921722412,
        "normalized": 0.9998935200270991,
        "contribution": 0.49994676001354954
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.207131385803223,
        "normalized": 0.989624923740222,
        "contribution": 0.1484437385610333
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.337693214416504,
        "normalized": 0.9542463921920685,
        "contribution": 0.09542463921920685
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "13f614e20538c124483a49f60f1dc19c8b23457233406996819f689f80dc1998",
    "id": "60c6875b6790876726d248c2e4fd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 3.616149425506592,
    "wer": 0.0,
    "identity": 0.6565101146697998,
    "voiceclap": 0.09229007363319397,
    "fitness": 0.7796236177959452,
    "legacy_fitness": 0.7796236177959452,
    "target_cdf": 0.9381054839846314,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.81038498878479,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7559464681909179,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.81038498878479,
        "normalized": 0.9381054839846314,
        "contribution": 0.4690527419923157
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2128617763519287,
        "normalized": 0.29357562439950946,
        "contribution": 0.07339390609987737
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4707932472229004,
        "normalized": 0.8206823740743052,
        "contribution": 0.08206823740743052
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0276e5bbbda05e5334df286e92078b3485f358b78506a370faa78e1123d1925f",
    "id": "612731a81b013f2a4a6130c0ba7d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.586543083190918,
    "genuineness": 4.381826400756836,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.02135176584124565,
    "fitness": 0.830934781605057,
    "legacy_fitness": 0.830934781605057,
    "target_cdf": 0.9673902089874984,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.126917839050293,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6830518399271794,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.126917839050293,
        "normalized": 0.9673902089874984,
        "contribution": 0.4836951044937492
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4427849054336548,
        "normalized": 0.07640335680063054,
        "contribution": 0.019100839200157635
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.694857597351074,
        "normalized": 0.8375344060053678,
        "contribution": 0.12563016090080517
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.567589521408081,
        "normalized": 0.5462573533246747,
        "contribution": 0.05462573533246748
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "47f1afb6508009f38b8d14c2656f6921fbf5cf653358b0011f0c7751b983f336",
    "id": "616e510bdff442142002df61545f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.157639026641846,
    "genuineness": 4.532451152801514,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": -0.04344230890274048,
    "fitness": 0.6222000272009144,
    "legacy_fitness": 0.6222000272009144,
    "target_cdf": 0.6017241870397882,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.4793236255645752,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5317654301957342,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.4793236255645752,
        "normalized": 0.6017241870397882,
        "contribution": 0.3008620935198941
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5571913719177246,
        "normalized": 0.44176949523922726,
        "contribution": 0.11044237380980682
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.334923028945923,
        "normalized": 0.4168490496018132,
        "contribution": 0.06252735744027198
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6622002124786377,
        "normalized": 0.5793360542576129,
        "contribution": 0.05793360542576129
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "01385b6dfbfd05b8063ae83249a83ee0a120d1059212e5f59a53b43b73bd5d72",
    "id": "620b59f7949125444a7b1af9a88f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": 0.631548285484314,
    "voiceclap": 0.036559950560331345,
    "fitness": 0.8261424235956261,
    "legacy_fitness": 0.8261424235956261,
    "target_cdf": 0.9995111649148442,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.636536121368408,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7824513241502522,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.636536121368408,
        "normalized": 0.9995111649148442,
        "contribution": 0.4997555824574221
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8736209869384766,
        "normalized": 0.17525676081671665,
        "contribution": 0.04381419020417916
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.771240711212158,
        "normalized": 0.9731772088639971,
        "contribution": 0.14597658132959956
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.091264724731445,
        "normalized": 0.9290497015905134,
        "contribution": 0.09290497015905134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5dd126d23a8d423e70aa4eeb4d8f87221ee778745794b3233d825ac5bb10b551",
    "id": "623aebd2cf2df069fe4facddfc61",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.499174118041992,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.028122998774051666,
    "fitness": 0.7774067679729755,
    "legacy_fitness": 0.7774067679729755,
    "target_cdf": 0.8824847618116329,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4451546669006348,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6955371770433324,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4451546669006348,
        "normalized": 0.8824847618116329,
        "contribution": 0.44124238090581647
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1144819259643555,
        "normalized": 0.25583161582232095,
        "contribution": 0.06395790395558024
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3568201065063477,
        "normalized": 0.7541161908392326,
        "contribution": 0.11311742862588489
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2771711349487305,
        "normalized": 0.7721946355605077,
        "contribution": 0.07721946355605078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c71112792f2023227e0d3f1da15283f9ab6726a899d4a0e51171dc44f65fe168",
    "id": "62471a03e91b3b8ddd3505784fd6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.823258876800537,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07534603029489517,
    "fitness": 0.8173410937489691,
    "legacy_fitness": 0.8173410937489691,
    "target_cdf": 0.9998555193944024,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.977417945861816,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8468345247228064,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.977417945861816,
        "normalized": 0.9998555193944024,
        "contribution": 0.4999277596972012
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.457115411758423,
        "normalized": 0.3967691763438149,
        "contribution": 0.09919229408595373
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.634126663208008,
        "normalized": 0.9964005462960434,
        "contribution": 0.1494600819444065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.812947750091553,
        "normalized": 0.9825438899524496,
        "contribution": 0.09825438899524497
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d5a5b30e1ff5930a274d7f202fd21ff127f2c142d2c1b2654c89cde0e7f0b7c4",
    "id": "6257acab06e74df952c70bea54e7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": 0.8255491256713867,
    "voiceclap": 0.021967804059386253,
    "fitness": 0.8040858846323837,
    "legacy_fitness": 0.8040858846323837,
    "target_cdf": 0.9706627967789073,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1756153106689453,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7041492883390693,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1756153106689453,
        "normalized": 0.9706627967789073,
        "contribution": 0.48533139838945366
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4088608026504517,
        "normalized": 0.0709502961187552,
        "contribution": 0.0177375740296888
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.917250871658325,
        "normalized": 0.8809377370125135,
        "contribution": 0.13214066055187704
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9933383464813232,
        "normalized": 0.6893965536804982,
        "contribution": 0.06893965536804983
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1c235b69747f0dbb9762093d229d3642198b412c53d90385f47350b27a2f6275",
    "id": "627c201c471d6a36666149397ffe",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.515059471130371,
    "genuineness": 4.137060642242432,
    "wer": 0.0,
    "identity": 0.7230312824249268,
    "voiceclap": -0.1288895606994629,
    "fitness": 0.780988218297634,
    "legacy_fitness": 0.780988218297634,
    "target_cdf": 0.8914684095185468,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.493852138519287,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6403147085448054,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.493852138519287,
        "normalized": 0.8914684095185468,
        "contribution": 0.4457342047592734
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2322797775268555,
        "normalized": 0.7184101418133952,
        "contribution": 0.10776152127200929
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.636897325515747,
        "normalized": 0.5705341849730415,
        "contribution": 0.05705341849730416
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8917e268880ece638939e1476f89d29f235cd376398890502d9441129b604dd0",
    "id": "62b9a79f38f9f0bf8323e7c7971d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.936227321624756,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -5.978718400001526e-05,
    "fitness": 0.8124562536571533,
    "legacy_fitness": 0.8124562536571533,
    "target_cdf": 0.9755090862302205,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.256777286529541,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7617967455391765,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.256777286529541,
        "normalized": 0.9755090862302205,
        "contribution": 0.48775454311511024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9312918186187744,
        "normalized": 0.19291961867711876,
        "contribution": 0.04822990466927969
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.255288600921631,
        "normalized": 0.9301035297844544,
        "contribution": 0.13951552946766815
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.668816089630127,
        "normalized": 0.8629676828711841,
        "contribution": 0.08629676828711841
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7dbecef2e91b6351d6340d24e80df46b0fc7434d7bdf162d18c2205e8102ae99",
    "id": "62e338cfd8c3841d04c71c0824ae",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.848362922668457,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.06127091124653816,
    "fitness": 0.8181637929463474,
    "legacy_fitness": 0.8181637929463474,
    "target_cdf": 0.9990117801396228,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.425514221191406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8764217124053233,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.425514221191406,
        "normalized": 0.9990117801396228,
        "contribution": 0.4995058900698114
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.7183308601379395,
        "normalized": 0.5155797250883145,
        "contribution": 0.12889493127207863
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.456212043762207,
        "normalized": 0.9943171611150623,
        "contribution": 0.14914757416725935
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 5.006569862365723,
        "normalized": 0.9887331689617396,
        "contribution": 0.09887331689617396
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "40afb8a2d649f373fa70471f02a0e3f9d9240737e48750148df59470f67d92cc",
    "id": "635f3fdbd7b175de45fc447cf2a9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.21237325668335,
    "wer": 0.0,
    "identity": 0.5045585632324219,
    "voiceclap": -0.02595607191324234,
    "fitness": 0.833077466122006,
    "legacy_fitness": 0.833077466122006,
    "target_cdf": 0.9985049319896988,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.295654773712158,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7699312517837812,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.295654773712158,
        "normalized": 0.9985049319896988,
        "contribution": 0.4992524659948494
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.637804985046387,
        "normalized": 0.965045519092528,
        "contribution": 0.1447568278638792
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9944534301757812,
        "normalized": 0.9166388033350146,
        "contribution": 0.09166388033350147
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "86fa8d964959e5423f149a3f9dbc5ca6bb712371fcf01f81e5f126c9f4974a58",
    "id": "6412b1f472b062ff6d045c2d418b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.76,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 3.8420867919921875,
    "wer": 0.0,
    "identity": 0.6444576978683472,
    "voiceclap": 0.08457627147436142,
    "fitness": 0.8020101984012105,
    "legacy_fitness": 0.8020101984012105,
    "target_cdf": 0.961270594530896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.045755386352539,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.713225327846368,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.045755386352539,
        "normalized": 0.961270594530896,
        "contribution": 0.480635297265448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6124054193496704,
        "normalized": 0.10853160036418047,
        "contribution": 0.027132900091045117
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9439380168914795,
        "normalized": 0.8855360174979171,
        "contribution": 0.13283040262468757
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.114352226257324,
        "normalized": 0.7262672786518737,
        "contribution": 0.07262672786518737
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1640e3f5182a4ffe1bb10b95a7c5e141166554a9b74eea7284769b641ca01341",
    "id": "6417ec33f3d2c5a0ec3e251c3ae6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.986435890197754,
    "wer": 0.0,
    "identity": 0.660672664642334,
    "voiceclap": 0.05666045472025871,
    "fitness": 0.819770664422427,
    "legacy_fitness": 0.819770664422427,
    "target_cdf": 0.9867481280341037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5164966583251953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7260013398099623,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5164966583251953,
        "normalized": 0.9867481280341037,
        "contribution": 0.49337406401705186
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5581269264221191,
        "normalized": 0.09733711108070414,
        "contribution": 0.024334277770176035
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.219705581665039,
        "normalized": 0.9258149280138626,
        "contribution": 0.1388722392020794
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.008740186691284,
        "normalized": 0.6942075882065499,
        "contribution": 0.069420758820655
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7a5e09e2dd6ab219f5d7cdc8c17c62e2cc74110ecd6315d95f40379823b7ac3c",
    "id": "6436271e80dc3b5e0ddf665b1389",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.049295634031295776,
    "fitness": 0.8050641139362318,
    "legacy_fitness": 0.8050641139362318,
    "target_cdf": 0.9585742155738795,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0132906436920166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7256752969535595,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0132906436920166,
        "normalized": 0.9585742155738795,
        "contribution": 0.4792871077869397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.086269855499268,
        "normalized": 0.9079316126330063,
        "contribution": 0.13618974189495092
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3079745769500732,
        "normalized": 0.780368248743694,
        "contribution": 0.07803682487436941
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "360e9d354d8996b5044b1fdb412e77858168f8c731cb285116debf5786206db4",
    "id": "654bb5e8a10a221a1722a6fd308d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": 0.743449330329895,
    "voiceclap": 0.04831458628177643,
    "fitness": 0.7768901896998888,
    "legacy_fitness": 0.7768901896998888,
    "target_cdf": 0.9079315965425183,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5912468433380127,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6845730531859641,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5912468433380127,
        "normalized": 0.9079315965425183,
        "contribution": 0.45396579827125916
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8702285289764404,
        "normalized": 0.17425071298327421,
        "contribution": 0.043562678245818554
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5436301231384277,
        "normalized": 0.8027714861707109,
        "contribution": 0.12041572292560664
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9207301139831543,
        "normalized": 0.6662885374327974,
        "contribution": 0.06662885374327975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7e0342b6332bb95ae6ddc28a23bc7127ebacedd6d02e9f452b1d45f4172c35f8",
    "id": "65ac3ac7a36ac3b09d4a614e26b7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.143336772918701,
    "wer": 0.0,
    "identity": 0.7743247747421265,
    "voiceclap": 0.03164658695459366,
    "fitness": 0.8274721956378259,
    "legacy_fitness": 0.8274721956378259,
    "target_cdf": 0.9950331004711068,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.889843463897705,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7549602966179227,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.889843463897705,
        "normalized": 0.9950331004711068,
        "contribution": 0.4975165502355534
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7752411365509033,
        "normalized": 0.1475786358230735,
        "contribution": 0.03689465895576838
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5059971809387207,
        "normalized": 0.8287452780397044,
        "contribution": 0.08287452780397045
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "713f03815a32db4eab41fda9dc768dcabbc622308fff125a8994aef12d09a055",
    "id": "65b7c999f7c20d05c37c389b5a15",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 4.3504462242126465,
    "wer": 0.0,
    "identity": 0.6713005304336548,
    "voiceclap": -0.18117253482341766,
    "fitness": 0.7832932510878673,
    "legacy_fitness": 0.7832932510878673,
    "target_cdf": 0.895770231665379,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5182008743286133,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6681346124950862,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5182008743286133,
        "normalized": 0.895770231665379,
        "contribution": 0.4478851158326895
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.043241500854492,
        "normalized": 0.23017181441662055,
        "contribution": 0.057542953604155136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.793053150177002,
        "normalized": 0.5762788231932616,
        "contribution": 0.08644182347898924
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.241966962814331,
        "normalized": 0.762647195792524,
        "contribution": 0.0762647195792524
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f6da58a6946dcb1a2a5478589ee2989c6663810b6decaacdb7a2874ee7b4087b",
    "id": "65dc37d900465170aaca56416552",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0325925350189209,
    "fitness": 0.8151615057600192,
    "legacy_fitness": 0.8151615057600192,
    "target_cdf": 0.9599408456112248,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0295231342315674,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7464387689184016,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0295231342315674,
        "normalized": 0.9599408456112248,
        "contribution": 0.4799704228056124
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9787856340408325,
        "normalized": 0.2082523969786677,
        "contribution": 0.05206309924466693
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5103979110717773,
        "normalized": 0.8297366417682784,
        "contribution": 0.08297366417682785
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "70c2548d1fd6bf989349831f061a3f41de1eb8615f38dc345b3e26c725d96a26",
    "id": "66555f74ad5cda0bd3d9a5fa2827",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10455454140901566,
    "fitness": 0.7874513962195168,
    "legacy_fitness": 0.7874513962195168,
    "target_cdf": 0.880937722986112,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4370384216308594,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6330876825493399,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4370384216308594,
        "normalized": 0.880937722986112,
        "contribution": 0.440468861493056
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6124054193496704,
        "normalized": 0.10853160036418047,
        "contribution": 0.027132900091045117
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4724645614624023,
        "normalized": 0.7849622943953769,
        "contribution": 0.11774434415930653
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.3728671073913574,
        "normalized": 0.4774157680593222,
        "contribution": 0.04774157680593222
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "40ecc123ee010fc2e7e4d4cef2e321f96db8456e612f165b2b42f2845b972d65",
    "id": "66f607cd19ee21159cc2ded57813",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.961331605911255,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07666610181331635,
    "fitness": 0.7944778836110518,
    "legacy_fitness": 0.7944778836110518,
    "target_cdf": 0.9418287707049221,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8428497314453125,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6919591715918698,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8428497314453125,
        "normalized": 0.9418287707049221,
        "contribution": 0.47091438535246105
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8464816808700562,
        "normalized": 0.16731154547607663,
        "contribution": 0.04182788636901916
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4368815422058105,
        "normalized": 0.7757183877941347,
        "contribution": 0.1163577581691202
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.80631685256958,
        "normalized": 0.6285914170126945,
        "contribution": 0.06285914170126945
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d10bcffa9b28d9a2f3b8a5f782f7cdd14b406f572053f6d476632f4ca2146bb7",
    "id": "6722bcec96c039331aaf01188050",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.557555198669434,
    "wer": 0.03571428571428571,
    "identity": 0.6183507442474365,
    "voiceclap": -0.18491366505622864,
    "fitness": 0.7265531081214991,
    "legacy_fitness": 0.7265531081214991,
    "target_cdf": 0.8144811411783466,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.140796184539795,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6070085509681438,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.140796184539795,
        "normalized": 0.8144811411783466,
        "contribution": 0.4072405705891733
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0194945335388184,
        "normalized": 0.22194973203149376,
        "contribution": 0.05548743300787344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.75524640083313,
        "normalized": 0.5632363808968689,
        "contribution": 0.08448545713453033
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.716106414794922,
        "normalized": 0.5979509023656668,
        "contribution": 0.059795090236566685
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4fa226058f0421ee5c96c86c78f8004abef15e083a4a740c8d86589afee590f6",
    "id": "67ca19bc36b781f0767d9ada2281",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.570107460021973,
    "wer": 0.0,
    "identity": 0.4643211364746094,
    "voiceclap": -0.20247898995876312,
    "fitness": 0.8321410909808221,
    "legacy_fitness": 0.8321410909808221,
    "target_cdf": 0.9644381951325223,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.086336612701416,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7075047738791924,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.086336612701416,
        "normalized": 0.9644381951325223,
        "contribution": 0.48221909756626113
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9787856340408325,
        "normalized": 0.2082523969786677,
        "contribution": 0.05206309924466693
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.321237087249756,
        "normalized": 0.744168371280294,
        "contribution": 0.1116252556920441
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7689125537872314,
        "normalized": 0.6159732137622032,
        "contribution": 0.061597321376220315
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "59676f418f0ae97b23014dfd182c3ce4a9a1e2ad9849c3a2384e304154d1cc3e",
    "id": "68032f12d24f24d1d0a31207a4e3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.229123115539551,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": 0.575514554977417,
    "voiceclap": -0.10167230665683746,
    "fitness": 0.7676529069177219,
    "legacy_fitness": 0.7676529069177219,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6183142019151552,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0194945335388184,
        "normalized": 0.22194973203149376,
        "contribution": 0.05548743300787344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.7941651344299316,
        "normalized": 0.576661242199227,
        "contribution": 0.08649918632988404
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.439974784851074,
        "normalized": 0.5011687918140428,
        "contribution": 0.050116879181404284
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ce104a41b0b978a7d822a1db9e5994db4c4c6392a5061508e4d5d021b8cff253",
    "id": "68b58ab11e2b3876b1a9ec4376dc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.3881025314331055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.051616549491882324,
    "fitness": 0.8335096948435712,
    "legacy_fitness": 0.8335096948435712,
    "target_cdf": 0.9811510787587977,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3704047203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7475170555850453,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3704047203063965,
        "normalized": 0.9811510787587977,
        "contribution": 0.49057553937939885
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0568110942840576,
        "normalized": 0.23494557063169752,
        "contribution": 0.05873639265792438
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5792131423950195,
        "normalized": 0.8113311992293513,
        "contribution": 0.12169967988440268
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.250767946243286,
        "normalized": 0.765054436633194,
        "contribution": 0.0765054436633194
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ee817d57a5547b5ba5505564d6e6c249c23e89e8d84687d19c5a37596b6e6257",
    "id": "68ebdf6dd093885d515a3e3eed54",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 4.7458367347717285,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0116199292242527,
    "fitness": 0.7812705342128933,
    "legacy_fitness": 0.7812705342128933,
    "target_cdf": 0.8559956519117078,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3152952194213867,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6218992048520703,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3152952194213867,
        "normalized": 0.8559956519117078,
        "contribution": 0.4279978259558539
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9245071411132812,
        "normalized": 0.1907869738584032,
        "contribution": 0.0476967434646008
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0855002403259277,
        "normalized": 0.6733630714845132,
        "contribution": 0.10100446072267698
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.3008086681365967,
        "normalized": 0.45200174708938656,
        "contribution": 0.04520017470893866
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fc4d3a05e2e1d8927389291d713926a28a6745eb33070cff131aa098d62ff084",
    "id": "68feaceea44d036b3615e52b9e36",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.262581825256348,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.14397428929805756,
    "fitness": 0.7615760067725533,
    "legacy_fitness": 0.7615760067725533,
    "target_cdf": 0.8469494938104061,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2747139930725098,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6221661953817097,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2747139930725098,
        "normalized": 0.8469494938104061,
        "contribution": 0.42347474690520304
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.174457550048828,
        "normalized": 0.7010225607218077,
        "contribution": 0.10515338410827114
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7381091117858887,
        "normalized": 0.6054879932877197,
        "contribution": 0.060548799328771975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3a2eb8cf9e3f93578c4dc94d591c1f8a78c163023913b6fbdb24c55a95667a32",
    "id": "69b4219bc9be3f1c4a5bbb1c40de",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.080576419830322,
    "wer": 0.0,
    "identity": 0.5917619466781616,
    "voiceclap": 0.01576262153685093,
    "fitness": 0.8195631561402532,
    "legacy_fitness": 0.8195631561402532,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7356690133427061,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.395984649658203,
        "normalized": 0.8027715080715057,
        "contribution": 0.08027715080715057
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a3614853cb5bcfc16bcca65c86456d78503bc100eefadd4528e2e15cfaa6db31",
    "id": "6a4d40be9240122a96ffc180251a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.099404335021973,
    "wer": 0.0,
    "identity": 0.609171986579895,
    "voiceclap": 0.053946204483509064,
    "fitness": 0.8288294490578852,
    "legacy_fitness": 0.8288294490578852,
    "target_cdf": 0.9994229746665722,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.587838649749756,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8083644355730384,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.587838649749756,
        "normalized": 0.9994229746665722,
        "contribution": 0.4997114873332861
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1823301315307617,
        "normalized": 0.2815898456276642,
        "contribution": 0.07039746140691605
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.869093894958496,
        "normalized": 0.9780804006212112,
        "contribution": 0.14671206009318166
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.985652446746826,
        "normalized": 0.9154342673965461,
        "contribution": 0.09154342673965461
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b081a5f671273f48e59542e3941ef809eeed101965fc4b4a23f2eab3f4d813ae",
    "id": "6a752491cbcb8e695fd0ebb17369",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": 0.6546264886856079,
    "voiceclap": 0.04074117913842201,
    "fitness": 0.7763860302431966,
    "legacy_fitness": 0.7763860302431966,
    "target_cdf": 0.8840175146388185,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.45327091217041,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6836570136139055,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.45327091217041,
        "normalized": 0.8840175146388185,
        "contribution": 0.44200875731940925
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0161020755767822,
        "normalized": 0.22078898327927737,
        "contribution": 0.05519724581981934
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.383507251739502,
        "normalized": 0.7614385460618895,
        "contribution": 0.11421578190928343
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1011507511138916,
        "normalized": 0.7223522856539355,
        "contribution": 0.07223522856539355
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "230f350dcadc695dc76cadb6a061e610aa02a6ac7e60f340a386287ae308ba1c",
    "id": "6b348fee7d9935f1559409ac6981",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04210175573825836,
    "fitness": 0.7973265465199456,
    "legacy_fitness": 0.7973265465199456,
    "target_cdf": 0.9269054842192939,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.72110652923584,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7346683632966996,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.72110652923584,
        "normalized": 0.9269054842192939,
        "contribution": 0.46345274210964693
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1144819259643555,
        "normalized": 0.25583161582232095,
        "contribution": 0.06395790395558024
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.730440616607666,
        "normalized": 0.845096013305189,
        "contribution": 0.12676440199577835
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.404785633087158,
        "normalized": 0.8049331523569412,
        "contribution": 0.08049331523569413
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4bace7b2c191ce95113195132255a727d0aeb300d4ce83e17fccc804a0aab748",
    "id": "6b751c3b79f106c1029bf9822a9a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.031691450625658035,
    "fitness": 0.8253767722555735,
    "legacy_fitness": 0.8253767722555735,
    "target_cdf": 0.9979798622347391,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1982598304748535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7806315839297168,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1982598304748535,
        "normalized": 0.9979798622347391,
        "contribution": 0.49898993111736956
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.917722225189209,
        "normalized": 0.18866878339787096,
        "contribution": 0.04716719584946774
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.637804985046387,
        "normalized": 0.965045519092528,
        "contribution": 0.1447568278638792
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.862438201904297,
        "normalized": 0.897176290990003,
        "contribution": 0.0897176290990003
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b33bbf3b334158b546a313675cb10ca65d0901ff9b7f4e022d491586703ef56e",
    "id": "6b8394bd5bd9440dab91e5bd4f33",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.76,
    "cfg": 3,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 3.553389072418213,
    "wer": 0.0,
    "identity": 0.7856491804122925,
    "voiceclap": 0.024859977886080742,
    "fitness": 0.7878438896094073,
    "legacy_fitness": 0.7878438896094073,
    "target_cdf": 0.9685130012341955,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1431503295898438,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7614397478812754,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1431503295898438,
        "normalized": 0.9685130012341955,
        "contribution": 0.48425650061709774
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9245071411132812,
        "normalized": 0.1907869738584032,
        "contribution": 0.0476967434646008
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.424307346343994,
        "normalized": 0.9479187150078625,
        "contribution": 0.14218780725117938
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7216219902038574,
        "normalized": 0.8729869654839748,
        "contribution": 0.08729869654839749
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b30a73979f6f73ebb3be24775f5170b66e6ba7ce36558a75e05da5b9df749eb4",
    "id": "6c271bdc53324b5f92919d43ae92",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.1020402908325195,
    "genuineness": 3.8295347690582275,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.02797708846628666,
    "fitness": 0.8119503015026603,
    "legacy_fitness": 0.8119503015026603,
    "target_cdf": 0.9996706910421757,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.750163555145264,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8206015578785945,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.750163555145264,
        "normalized": 0.9996706910421757,
        "contribution": 0.4998353455210878
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.219646692276001,
        "normalized": 0.2962709117111364,
        "contribution": 0.0740677279277841
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.456212043762207,
        "normalized": 0.9943171611150623,
        "contribution": 0.14914757416725935
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.654529571533203,
        "normalized": 0.9755091026246321,
        "contribution": 0.09755091026246321
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "81a81d302898ab87aa323d51ea2b0af2a905602a19fd7a83786e9b4960202419",
    "id": "6c64920eeec5254624d051290063",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 3.7291181087493896,
    "wer": 0.0,
    "identity": 0.7701506018638611,
    "voiceclap": 0.10974390059709549,
    "fitness": 0.8096804809000397,
    "legacy_fitness": 0.8096804809000397,
    "target_cdf": 0.997542097454529,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.13332986831665,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8274225363769625,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.13332986831665,
        "normalized": 0.997542097454529,
        "contribution": 0.4987710487272645
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.304456949234009,
        "normalized": 0.33087440203678953,
        "contribution": 0.08271860050919738
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.402837753295898,
        "normalized": 0.9935110498860812,
        "contribution": 0.14902665748291216
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.540116310119629,
        "normalized": 0.9690622965758837,
        "contribution": 0.09690622965758838
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0d78927a569b7b2ec7b17e88e043205b7c29f7176e651d69f2e8fd47f96fec71",
    "id": "6c8dde93cc4cba3811e024bfc687",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": 0.6433231830596924,
    "voiceclap": -0.06943699717521667,
    "fitness": 0.7635067088285925,
    "legacy_fitness": 0.7635067088285925,
    "target_cdf": 0.8713534639040315,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.388340950012207,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6599626191557866,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.388340950012207,
        "normalized": 0.8713534639040315,
        "contribution": 0.43567673195201573
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8091652393341064,
        "normalized": 0.1567722644593298,
        "contribution": 0.03919306611483245
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4546730518341064,
        "normalized": 0.7803682683335784,
        "contribution": 0.11705524025003676
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9647350311279297,
        "normalized": 0.6803758083890165,
        "contribution": 0.06803758083890164
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4a735e31d3922198e8738ad71018dc536404892574901bec4bdaf04909f587c8",
    "id": "6cd14b8bedcb229adc6f621ef881",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.20609712600708,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.15438836812973022,
    "fitness": 0.8309206424280848,
    "legacy_fitness": 0.8309206424280848,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7219092030238491,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5445572137832642,
        "normalized": 0.09467573940569673,
        "contribution": 0.023668934851424184
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.193018436431885,
        "normalized": 0.9224687261883889,
        "contribution": 0.13837030892825833
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8019163608551025,
        "normalized": 0.6271137841199856,
        "contribution": 0.06271137841199857
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4ed7802fb41943d215c7bf680e2db53ec7213c0309661552852b036a8f0eafb8",
    "id": "6d30613059aebd395c9194a96ca4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.5220088958740234,
    "wer": 0.0,
    "identity": 0.7282232046127319,
    "voiceclap": 0.010666266083717346,
    "fitness": 0.7971806578241502,
    "legacy_fitness": 0.7971806578241502,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7291603396061713,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6191902160644531,
        "normalized": 0.1099935309435226,
        "contribution": 0.02749838273588065
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.121852874755859,
        "normalized": 0.9129861397194918,
        "contribution": 0.13694792095792377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.153956890106201,
        "normalized": 0.7378459799661322,
        "contribution": 0.07378459799661323
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1bd2941e78e348ae25843eb120391e92fd3fdf52752b83a5e8f708be3239ced0",
    "id": "6d6cd0eb2d0621bad9b4a605ce58",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.582659721374512,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.19243121147155762,
    "fitness": 0.7351016582929221,
    "legacy_fitness": 0.7351016582929221,
    "target_cdf": 0.7780502521759524,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.00687837600708,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5926662732042476,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.00687837600708,
        "normalized": 0.7780502521759524,
        "contribution": 0.3890251260879762
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.124659299850464,
        "normalized": 0.2596155287980314,
        "contribution": 0.06490388219950785
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8475394248962402,
        "normalized": 0.5949256955457561,
        "contribution": 0.08923885433186342
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4225101470947266,
        "normalized": 0.4949841058490014,
        "contribution": 0.04949841058490014
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f9b1fcce2fabd5af151cac22eb24ecf736dc5b9f3e7bf6abb2c42d209a9d166a",
    "id": "6d8a4393d9ebbd264669197f43bd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.05311785265803337,
    "fitness": 0.825211850177123,
    "legacy_fitness": 0.825211850177123,
    "target_cdf": 0.9979798622347391,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1982598304748535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.822453526561006,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1982598304748535,
        "normalized": 0.9979798622347391,
        "contribution": 0.49898993111736956
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2891910076141357,
        "normalized": 0.3245264899668597,
        "contribution": 0.08113162249171492
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.886885643005371,
        "normalized": 0.97888519494114,
        "contribution": 0.146832779241171
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.346494197845459,
        "normalized": 0.9549919371075053,
        "contribution": 0.09549919371075054
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "be39c0a2d8f177e04b80d8ebaf83298468cf04ce41c35e0be52ce83310bd8eb6",
    "id": "6d9157c12880e16e9c4b8bb6d281",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": 0.6785023212432861,
    "voiceclap": -0.1275871992111206,
    "fitness": 0.7704931633247816,
    "legacy_fitness": 0.7704931633247816,
    "target_cdf": 0.8729869637040064,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3964571952819824,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6301137994880176,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3964571952819824,
        "normalized": 0.8729869637040064,
        "contribution": 0.4364934818520032
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.632759928703308,
        "normalized": 0.11295969157872005,
        "contribution": 0.02823992289468001
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.125530958175659,
        "normalized": 0.6859401050066158,
        "contribution": 0.10289101575099237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7953155040740967,
        "normalized": 0.6248937899034204,
        "contribution": 0.06248937899034204
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0d693916a4e21ec3e97e3500dd358b9a8d7d48449921b2b9796e5b90d5c1c112",
    "id": "6da3848e33881b8fb364ee28c180",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": 0.49384865164756775,
    "voiceclap": -0.07872240990400314,
    "fitness": 0.8144878944706984,
    "legacy_fitness": 0.8144878944706984,
    "target_cdf": 0.9527255219125881,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9483609199523926,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.727252243945523,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9483609199523926,
        "normalized": 0.9527255219125881,
        "contribution": 0.47636276095629404
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.005924940109253,
        "normalized": 0.21732775018194794,
        "contribution": 0.054331937545486986
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.588109016418457,
        "normalized": 0.8134348350041303,
        "contribution": 0.12201522525061953
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1803598403930664,
        "normalized": 0.745423201931225,
        "contribution": 0.0745423201931225
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e7aa95452671962817b5fd2cbfa0c1cfa6cce4ccb8c0c7dad1b62d523db42fe2",
    "id": "6da9a2916c8f28b0ab6bf00a3ee2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 3.6851859092712402,
    "wer": 0.0,
    "identity": 0.5168343782424927,
    "voiceclap": 0.029908714815974236,
    "fitness": 0.803657357082869,
    "legacy_fitness": 0.803657357082869,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.72975489964564,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5716965198516846,
        "normalized": 0.1000529371132764,
        "contribution": 0.0250132342783191
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.250767946243286,
        "normalized": 0.765054436633194,
        "contribution": 0.0765054436633194
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "208865ff67c4a248add655993dcb79bcce4540c5a1f9bf833ee5f5283dc76f23",
    "id": "6dd1ebd6114cea8a3d7950101322",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.9676074981689453,
    "wer": 0.0,
    "identity": 0.682609498500824,
    "voiceclap": -0.028315799310803413,
    "fitness": 0.8063143654054834,
    "legacy_fitness": 0.8063143654054834,
    "target_cdf": 0.9606102950148648,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0376393795013428,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7391550001035281,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0376393795013428,
        "normalized": 0.9606102950148648,
        "contribution": 0.4803051475074324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9584312438964844,
        "normalized": 0.20159505719328874,
        "contribution": 0.050398764298322185
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3255765438079834,
        "normalized": 0.7849622466803536,
        "contribution": 0.07849622466803537
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3a5a7d06ced217dd66d46311ff0d665d78f3929453d62e0bf43cfa91ff6a4376",
    "id": "6edd36ce95c242737446f7974001",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 3.578493118286133,
    "wer": 0.0,
    "identity": 0.5605907440185547,
    "voiceclap": 0.07003682851791382,
    "fitness": 0.8007342910890011,
    "legacy_fitness": 0.8007342910890011,
    "target_cdf": 0.9937903356798943,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.808681011199951,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7941160130718881,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.808681011199951,
        "normalized": 0.9937903356798943,
        "contribution": 0.49689516783994714
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.03645658493042,
        "normalized": 0.2278053936557905,
        "contribution": 0.056951348413947625
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.833510875701904,
        "normalized": 0.9763929720149331,
        "contribution": 0.14645894580223998
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.170474052429199,
        "normalized": 0.9381055101575337,
        "contribution": 0.09381055101575338
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9d17a406f64b2278ac1ac508d5a09a7dde8f0594544a1e0a449bb537b50bab52",
    "id": "6fabf68d000077d516fd1cd6ddc6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": 0.5278651714324951,
    "voiceclap": -0.07473438233137131,
    "fitness": 0.7020455073692584,
    "legacy_fitness": 0.7020455073692584,
    "target_cdf": 0.7429102562622595,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.889193058013916,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5857087883526177,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.889193058013916,
        "normalized": 0.7429102562622595,
        "contribution": 0.37145512813112974
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0500261783599854,
        "normalized": 0.23255183995970719,
        "contribution": 0.058137959989926796
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8920180797576904,
        "normalized": 0.6099914805911717,
        "contribution": 0.09149872208867575
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8591229915618896,
        "normalized": 0.6461697814288547,
        "contribution": 0.06461697814288547
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b3682bb2f4bc8e274f1bec3a47c651623136569615d4f4e7f788232c8ba357a3",
    "id": "6fb2755955cfd3567362eea7d7b4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.538887023925781,
    "genuineness": 4.250029563903809,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.02793743833899498,
    "fitness": 0.8165317592848257,
    "legacy_fitness": 0.8165317592848257,
    "target_cdf": 0.9519500370480449,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.940244674682617,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6840846250887294,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.940244674682617,
        "normalized": 0.9519500370480449,
        "contribution": 0.47597501852402246
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6463295221328735,
        "normalized": 0.11598248129215816,
        "contribution": 0.02899562032303954
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.365715742111206,
        "normalized": 0.7565702742161262,
        "contribution": 0.11348554113241893
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8899264335632324,
        "normalized": 0.6562844510924838,
        "contribution": 0.06562844510924838
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ef938b69ea0575d3623fb822932624c06e7c6e2552eec37b0cc45ed325b45eae",
    "id": "6fb85cd4f7fbd63308bf40c928c4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": 0.6051687002182007,
    "voiceclap": -0.0970078706741333,
    "fitness": 0.8258742013682219,
    "legacy_fitness": 0.8258742013682219,
    "target_cdf": 0.9736578860102679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2243127822875977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7399474713524709,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2243127822875977,
        "normalized": 0.9736578860102679,
        "contribution": 0.48682894300513396
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9787856340408325,
        "normalized": 0.2082523969786677,
        "contribution": 0.05206309924466693
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.730440616607666,
        "normalized": 0.845096013305189,
        "contribution": 0.12676440199577835
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1715588569641113,
        "normalized": 0.7429102710689157,
        "contribution": 0.07429102710689157
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f102949300cb86c7a282e5848774662631f0a64360e3914f018a8bc12f5558c8",
    "id": "707c88767ff3a22419853c17bf8d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 4.18099308013916,
    "wer": 0.06666666666666667,
    "identity": 0.3440746068954468,
    "voiceclap": -0.05385780707001686,
    "fitness": 0.7669402081623343,
    "legacy_fitness": 0.7669402081623343,
    "target_cdf": 0.9867481280341037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5164966583251953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7726393294747027,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5164966583251953,
        "normalized": 0.9867481280341037,
        "contribution": 0.49337406401705186
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0194945335388184,
        "normalized": 0.22194973203149376,
        "contribution": 0.05548743300787344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.442098617553711,
        "normalized": 0.9495623364922869,
        "contribution": 0.14243435047384304
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4399898052215576,
        "normalized": 0.8134348197593442,
        "contribution": 0.08134348197593443
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f9e096567cdec39c0422e935932b57e998ea0a2dae30414baec26e965e84c53b",
    "id": "70be647705d02fc1eb48f22bc109",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.3504462242126465,
    "wer": 0.0,
    "identity": 0.6967110633850098,
    "voiceclap": -0.07632080465555191,
    "fitness": 0.825326965892123,
    "legacy_fitness": 0.825326965892123,
    "target_cdf": 0.9726892523841197,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2080800533294678,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7052946062725086,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2080800533294678,
        "normalized": 0.9726892523841197,
        "contribution": 0.48634462619205987
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8295196294784546,
        "normalized": 0.16246558721860938,
        "contribution": 0.040616396804652344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3434762954711914,
        "normalized": 0.7504102103482133,
        "contribution": 0.11256153155223199
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.89432692527771,
        "normalized": 0.6577205172356444,
        "contribution": 0.06577205172356444
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "577125063b81b8f65532fde0f5be6ced71e4e443efb1f8abf6e572f809beceaa",
    "id": "70e1e0b1bc422eea82f492630ffd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 3.8420867919921875,
    "wer": 0.0,
    "identity": 0.6053515076637268,
    "voiceclap": 0.03227682411670685,
    "fitness": 0.7820390119124027,
    "legacy_fitness": 0.7820390119124027,
    "target_cdf": 0.9213282215532803,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.680525302886963,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7021074483329566,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.680525302886963,
        "normalized": 0.9213282215532803,
        "contribution": 0.46066411077664016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8804056644439697,
        "normalized": 0.1772798100559148,
        "contribution": 0.0443199525139787
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5658695697784424,
        "normalized": 0.8081485136597149,
        "contribution": 0.12122227704895722
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2287654876708984,
        "normalized": 0.7590110799338052,
        "contribution": 0.07590110799338053
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2ec1a106b2327fa162645d31759f45c8a6504f623dc7d0c29ddecf4bce6f2273",
    "id": "71159fb20bcc012dd4ed595a2187",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.9425034523010254,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06901667267084122,
    "fitness": 0.8151927596781099,
    "legacy_fitness": 0.8151927596781099,
    "target_cdf": 0.9796648197280255,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.337939739227295,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7280390752741362,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.337939739227295,
        "normalized": 0.9796648197280255,
        "contribution": 0.4898324098640128
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.58526611328125,
        "normalized": 0.10282367927847097,
        "contribution": 0.025705919819617742
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.272770404815674,
        "normalized": 0.7710130971983604,
        "contribution": 0.07710130971983604
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "dde3da9782e2f0d66b0e9b3bf990f364c509eec066ec7a76b5f2054f246eeaca",
    "id": "714802bb145fd32ac45551001357",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": 0.21001340448856354,
    "voiceclap": -0.09933227300643921,
    "fitness": 0.81309740251961,
    "legacy_fitness": 0.81309740251961,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8077140118706156,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.307849407196045,
        "normalized": 0.33229177143790456,
        "contribution": 0.08307294285947614
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.522160530090332,
        "normalized": 0.9564536832164738,
        "contribution": 0.14346805248247108
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.029657363891602,
        "normalized": 0.921328195271452,
        "contribution": 0.09213281952714521
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8c49d8372386b5dde8e8cc82ca40b4af906845bbfcaf8021a71e921cd8848d47",
    "id": "719ed080d661341e364104581ac0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": 0.7303652763366699,
    "voiceclap": 0.004336181096732616,
    "fitness": 0.7889389689862236,
    "legacy_fitness": 0.7889389689862236,
    "target_cdf": 0.9269054842192939,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.72110652923584,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7229523050119653,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.72110652923584,
        "normalized": 0.9269054842192939,
        "contribution": 0.46345274210964693
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.066988229751587,
        "normalized": 0.2385614708758641,
        "contribution": 0.059640367718966025
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.605900287628174,
        "normalized": 0.8175982108486767,
        "contribution": 0.12263973162730149
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2771711349487305,
        "normalized": 0.7721946355605077,
        "contribution": 0.07721946355605078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "638e3b88c4b2388f13221ce04d6cd65128e97ed3244df9cb15482210b335a0d4",
    "id": "71b449db2cf078bde344143f682b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04971384257078171,
    "fitness": 0.8127727093279056,
    "legacy_fitness": 0.8127727093279056,
    "target_cdf": 0.9967235449909552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.035935401916504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7776935506337308,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.035935401916504,
        "normalized": 0.9967235449909552,
        "contribution": 0.4983617724954776
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.84308922290802,
        "normalized": 0.16633496146433524,
        "contribution": 0.04158374036608381
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.8157196044921875,
        "normalized": 0.9755091035016348,
        "contribution": 0.14632636552524522
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.976851463317871,
        "normalized": 0.9142167224692412,
        "contribution": 0.09142167224692413
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5ab1e3f8bbd35ba7e2a339250d3285e4d1170123640f531145df8dff6dca8501",
    "id": "71e62639879159f5e61772138e91",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04688511788845062,
    "fitness": 0.825584881220056,
    "legacy_fitness": 0.825584881220056,
    "target_cdf": 0.9970202380692827,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.0684003829956055,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7742175857720599,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.0684003829956055,
        "normalized": 0.9970202380692827,
        "contribution": 0.4985101190346414
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.71786642074585,
        "normalized": 0.9701371067780673,
        "contribution": 0.1455205660167101
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7392241954803467,
        "normalized": 0.8762105466288791,
        "contribution": 0.08762105466288791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "52ce92d5d638214b6ea0733c2131528a964abf53af91f026925f34f99fb78bd5",
    "id": "72381a752b93a99891c204d80fc9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 3.653805732727051,
    "wer": 0.0,
    "identity": 0.7694253921508789,
    "voiceclap": 0.04387021064758301,
    "fitness": 0.8007648195327087,
    "legacy_fitness": 0.8007648195327087,
    "target_cdf": 0.9772498618564534,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2892422676086426,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7421008357022214,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2892422676086426,
        "normalized": 0.9772498618564534,
        "contribution": 0.4886249309282267
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.26418399810791,
        "normalized": 0.9311452345235685,
        "contribution": 0.13967178517853526
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.41798734664917,
        "normalized": 0.8081485455599589,
        "contribution": 0.0808148545559959
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6253a87de98e87b5b5ed90dc79a9368031f869b295b6c4ac10852ab7fa234a3f",
    "id": "72949cbe6509ebc0623a10e446da",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": 0.6289724707603455,
    "voiceclap": 0.010350104421377182,
    "fitness": 0.798934815966891,
    "legacy_fitness": 0.798934815966891,
    "target_cdf": 0.9726892523841197,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2080800533294678,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7288294940219291,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2080800533294678,
        "normalized": 0.9726892523841197,
        "contribution": 0.48634462619205987
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6124054193496704,
        "normalized": 0.10853160036418047,
        "contribution": 0.027132900091045117
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.139644145965576,
        "normalized": 0.9154342857630313,
        "contribution": 0.1373151428644547
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3079745769500732,
        "normalized": 0.780368248743694,
        "contribution": 0.07803682487436941
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "89aa9745178a33d8e42a75153741eabdc928b91647eb3310ddcdd5b7a979832e",
    "id": "72b6fc485d1b5c2b12931882eed4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.012883320450782776,
    "fitness": 0.8330617156081839,
    "legacy_fitness": 0.8330617156081839,
    "target_cdf": 0.9796648197280255,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.337939739227295,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7654302045230018,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.337939739227295,
        "normalized": 0.9796648197280255,
        "contribution": 0.4898324098640128
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1517984867095947,
        "normalized": 0.26984519049493305,
        "contribution": 0.06746129762373326
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.032895088195801,
        "normalized": 0.8999470264638179,
        "contribution": 0.13499205396957267
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1319541931152344,
        "normalized": 0.7314444306568307,
        "contribution": 0.07314444306568307
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8b7ae1a46e84d19a9e85e889ef154ef63ca0a9beafadc24e51702ad4c3da8250",
    "id": "72cbcd7f431d6364eae9888cb498",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.6705241203308105,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09386880695819855,
    "fitness": 0.7791253228417943,
    "legacy_fitness": 0.7791253228417943,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5979583057839281,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.7630300521850586,
        "normalized": 0.5659276688993419,
        "contribution": 0.08488915033490128
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.2749557495117188,
        "normalized": 0.44292605995212875,
        "contribution": 0.044292605995212876
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "293b87f1202a4def1e977056bf2437630786ee47195f8b4dcddb7cc611b18dc1",
    "id": "72d8badad205a703cb7ad7d59cc6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0200495682656765,
    "fitness": 0.7899122694843894,
    "legacy_fitness": 0.7899122694843894,
    "target_cdf": 0.9026628930794968,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.558781862258911,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6800033839782522,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.558781862258911,
        "normalized": 0.9026628930794968,
        "contribution": 0.4513314465397484
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.03645658493042,
        "normalized": 0.2278053936557905,
        "contribution": 0.056951348413947625
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1700096130371094,
        "normalized": 0.6996650705664804,
        "contribution": 0.10494976058497206
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.925130605697632,
        "normalized": 0.6677082843958411,
        "contribution": 0.06677082843958411
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "da1bd7a137f8a4a25e2f430153bd7b4eaaf47eda029566d78b8362c0eb999e89",
    "id": "73425b6283f976fa373f37761b9d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.607763767242432,
    "wer": 0.0,
    "identity": 0.5889156460762024,
    "voiceclap": -0.13304053246974945,
    "fitness": 0.8451087481373848,
    "legacy_fitness": 0.8451087481373848,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7029668402115323,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4088608026504517,
        "normalized": 0.0709502961187552,
        "contribution": 0.0177375740296888
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.792710781097412,
        "normalized": 0.8577606879337987,
        "contribution": 0.1286641031900698
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8327198028564453,
        "normalized": 0.6374167680329237,
        "contribution": 0.06374167680329237
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c59750f19fae81d432a0cd57d245e9b70145eaba96fb25a310d721d4a32712fa",
    "id": "7413445b27951987a42e691d864d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": 0.6468932628631592,
    "voiceclap": 0.03092750906944275,
    "fitness": 0.8364405370996171,
    "legacy_fitness": 0.8364405370996171,
    "target_cdf": 0.996049248624995,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.971005439758301,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7930206649092564,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.971005439758301,
        "normalized": 0.996049248624995,
        "contribution": 0.4980246243124975
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.195899724960327,
        "normalized": 0.28688766529430837,
        "contribution": 0.07172191632357709
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6336121559143066,
        "normalized": 0.8559956465055119,
        "contribution": 0.08559956465055119
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f2c7a5a843e264661870320642d385a580c57ceddcb712379286348063ec0a84",
    "id": "74146f155b3f3ad4b6f17983e5fb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 4.137060642242432,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.001125408336520195,
    "fitness": 0.8229261965294303,
    "legacy_fitness": 0.8229261965294303,
    "target_cdf": 0.9948039235154036,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8736109733581543,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7673308475225024,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8736109733581543,
        "normalized": 0.9948039235154036,
        "contribution": 0.4974019617577018
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8498740196228027,
        "normalized": 0.16829178263767364,
        "contribution": 0.04207294565941841
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.442098617553711,
        "normalized": 0.9495623364922869,
        "contribution": 0.14243435047384304
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6248111724853516,
        "normalized": 0.8542158963153907,
        "contribution": 0.08542158963153908
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6c01f943c1bef2b10e6d446276fc4898010748f3262321228332d3dce97677d3",
    "id": "74596c5750e41cbd385225d9ca73",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.149612903594971,
    "wer": 0.0,
    "identity": 0.505375862121582,
    "voiceclap": -0.1041213870048523,
    "fitness": 0.824401870791097,
    "legacy_fitness": 0.824401870791097,
    "target_cdf": 0.9832067001342365,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.419102191925049,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7372507036552326,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.419102191925049,
        "normalized": 0.9832067001342365,
        "contribution": 0.49160335006711825
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8939753770828247,
        "normalized": 0.18137005253541438,
        "contribution": 0.045342513133853596
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.632587432861328,
        "normalized": 0.8237335195645583,
        "contribution": 0.12356002793468374
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.259568929672241,
        "normalized": 0.7674481251957699,
        "contribution": 0.07674481251957699
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2cd50e110cf5aa63a8c787abb6719ba563b5c1fee5af7d8fda57b321d7c2df42",
    "id": "74c841ec89766c226fba9f0bd497",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 4.086852550506592,
    "wer": 0.0,
    "identity": 0.605138897895813,
    "voiceclap": 0.044360414147377014,
    "fitness": 0.8242782877326026,
    "legacy_fitness": 0.8242782877326026,
    "target_cdf": 0.9997064444303541,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.782628059387207,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7827670640301618,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.782628059387207,
        "normalized": 0.9997064444303541,
        "contribution": 0.49985322221517703
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8295196294784546,
        "normalized": 0.16246558721860938,
        "contribution": 0.040616396804652344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.064799785614014,
        "normalized": 0.9856469755282963,
        "contribution": 0.14784704632924445
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.232080936431885,
        "normalized": 0.9445039868108795,
        "contribution": 0.09445039868108795
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "038b75326136d05f79d70535058d226431c928570afc35cad104c955ad2c3401",
    "id": "755802d7ad58d6a53686315fde15",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": 0.5485272407531738,
    "voiceclap": -0.04269580915570259,
    "fitness": 0.8285685311495882,
    "legacy_fitness": 0.8285685311495882,
    "target_cdf": 0.9948039235154036,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8736109733581543,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7750421213110186,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8736109733581543,
        "normalized": 0.9948039235154036,
        "contribution": 0.4974019617577018
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.948253870010376,
        "normalized": 0.19831469683683772,
        "contribution": 0.04957867420920943
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.246392726898193,
        "normalized": 0.9290496824330181,
        "contribution": 0.1393574523649527
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8008313179016113,
        "normalized": 0.8870403297915466,
        "contribution": 0.08870403297915466
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cff33157a52424cb8555ed4cc04e3e32af880e9c7990338526c14f8bc9655c18",
    "id": "766b44f29223b5423c317dbd0598",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": 0.632319986820221,
    "voiceclap": 0.015993427485227585,
    "fitness": 0.7720944111186977,
    "legacy_fitness": 0.7720944111186977,
    "target_cdf": 0.9154342620847058,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.639944076538086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6939965321913453,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.639944076538086,
        "normalized": 0.9154342620847058,
        "contribution": 0.4577171310423529
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8668360710144043,
        "normalized": 0.17324834271715545,
        "contribution": 0.04331208567928886
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.490255832672119,
        "normalized": 0.7894999685105402,
        "contribution": 0.11842499527658101
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1803598403930664,
        "normalized": 0.745423201931225,
        "contribution": 0.0745423201931225
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "000c66828de3867d413bd21acf8c195de248f84efc35750b996325835f365928",
    "id": "767e8ae1b06083b357bc50da631f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.413206577301025,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.02390758879482746,
    "fitness": 0.7948551830375166,
    "legacy_fitness": 0.7948551830375166,
    "target_cdf": 0.905324255769247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.575014352798462,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6573245853611193,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.575014352798462,
        "normalized": 0.905324255769247,
        "contribution": 0.4526621278846235
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9991400241851807,
        "normalized": 0.2150377207338816,
        "contribution": 0.0537594301834704
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8041727542877197,
        "normalized": 0.580099640087578,
        "contribution": 0.0870149460131367
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.837120532989502,
        "normalized": 0.6388808127988868,
        "contribution": 0.06388808127988868
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2ead21dfd3c0a697fbbd2bacbf0e201d3f06a73dec0df42699ef97ab9fb9638a",
    "id": "76cbf247bcbe5affc264bda3d831",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04410085454583168,
    "fitness": 0.8326761523684993,
    "legacy_fitness": 0.8326761523684993,
    "target_cdf": 0.9887331723771762,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5814266204833984,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7212644651675234,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5814266204833984,
        "normalized": 0.9887331723771762,
        "contribution": 0.4943665861885881
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4970635175704956,
        "normalized": 0.08578327395609947,
        "contribution": 0.021445818489024868
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.201914310455322,
        "normalized": 0.9235966507707382,
        "contribution": 0.1385394976156107
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9295310974121094,
        "normalized": 0.6691256287429975,
        "contribution": 0.06691256287429975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "18650ecae84fab298538e6a1d75b6508c69eb251ccd7dd3bdf81b37a145165da",
    "id": "771bba12ddec90339a9dd599288d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": 0.7646312713623047,
    "voiceclap": 0.0004925895482301712,
    "fitness": 0.8269556174064637,
    "legacy_fitness": 0.8269556174064637,
    "target_cdf": 0.9980776169667565,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.214492321014404,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7698910714709655,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.214492321014404,
        "normalized": 0.9980776169667565,
        "contribution": 0.4990388084833782
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.424307346343994,
        "normalized": 0.9479187150078625,
        "contribution": 0.14218780725117938
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8184332847595215,
        "normalized": 0.890006464053272,
        "contribution": 0.0890006464053272
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e765536208808ed6a9ac6e1b96a4260872ba788cbc9c608dd9807cb92b5d5e6e",
    "id": "7733fb900a36b793e9c9a9217959",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 3.986435890197754,
    "wer": 0.0,
    "identity": 0.6420891880989075,
    "voiceclap": 0.046545859426259995,
    "fitness": 0.7647012302554363,
    "legacy_fitness": 0.7647012302554363,
    "target_cdf": 0.8778006373398928,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4208059310913086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6891734901263691,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4208059310913086,
        "normalized": 0.8778006373398928,
        "contribution": 0.4389003186699464
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0296716690063477,
        "normalized": 0.2254527264287831,
        "contribution": 0.05636318160719578
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.539182424545288,
        "normalized": 0.8016852878192227,
        "contribution": 0.1202527931728834
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1495563983917236,
        "normalized": 0.7365719667634352,
        "contribution": 0.07365719667634352
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "177d1e207c81b0a0ad10e8f78daa9e51d64c82528af09d458ff646c156424278",
    "id": "77430c29bfc763acf440b8ed826e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": 0.7027376294136047,
    "voiceclap": -0.020676154643297195,
    "fitness": 0.8292705864448966,
    "legacy_fitness": 0.8292705864448966,
    "target_cdf": 0.9948039235154036,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8736109733581543,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7468325066525169,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8736109733581543,
        "normalized": 0.9948039235154036,
        "contribution": 0.4974019617577018
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.979520797729492,
        "normalized": 0.8914683718087433,
        "contribution": 0.1337202557713115
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1319541931152344,
        "normalized": 0.7314444306568307,
        "contribution": 0.07314444306568307
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e337765c8fbd23ec1acb670ad55515622d95b5a50bc73b2330652499a34f8fe0",
    "id": "77bdc08ee04b168a9c64b09d17a3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": 0.5761270523071289,
    "voiceclap": -0.00010161474347114563,
    "fitness": 0.83203379883734,
    "legacy_fitness": 0.83203379883734,
    "target_cdf": 0.9981710548076552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.230724811553955,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7505043657471432,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.230724811553955,
        "normalized": 0.9981710548076552,
        "contribution": 0.4990855274038276
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6666839122772217,
        "normalized": 0.12062361006755057,
        "contribution": 0.030155902516887642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.290871620178223,
        "normalized": 0.9341987553238786,
        "contribution": 0.1401298132985818
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4311888217926025,
        "normalized": 0.8113312252784615,
        "contribution": 0.08113312252784616
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6579db18ec407af353da9a8c639cd3e7ab2c3a7a69961aeb25e00f420f492051",
    "id": "77ed0018f0633716dda5e7de0040",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": 0.592702329158783,
    "voiceclap": 0.01587487757205963,
    "fitness": 0.833487958739109,
    "legacy_fitness": 0.833487958739109,
    "target_cdf": 0.9972925007937494,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.100865364074707,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7908334883580879,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.100865364074707,
        "normalized": 0.9972925007937494,
        "contribution": 0.4986462503968747
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9991400241851807,
        "normalized": 0.2150377207338816,
        "contribution": 0.0537594301834704
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.67338752746582,
        "normalized": 0.9673901905993718,
        "contribution": 0.14510852858990578
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "90ef4c386b33c2434d254380b29b856a32192340571af89b00258b07e02bff91",
    "id": "783537a5bdf7cba779bd5b49e111",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07228497415781021,
    "fitness": 0.8334433253731368,
    "legacy_fitness": 0.8334433253731368,
    "target_cdf": 0.9950331004711068,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.889843463897705,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7152903584269943,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.889843463897705,
        "normalized": 0.9950331004711068,
        "contribution": 0.4975165502355534
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.252810001373291,
        "normalized": 0.049631459384851806,
        "contribution": 0.012407864846212951
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0263421535491943,
        "normalized": 0.6996650747455826,
        "contribution": 0.06996650747455826
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0a1331f4856fb91686c354740ca008122692ed3c6d4e9277d8eeb1de7f88c872",
    "id": "78788f510506575a9ee7f5fbb8f8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": 0.6520328521728516,
    "voiceclap": 0.08712932467460632,
    "fitness": 0.8125036559640642,
    "legacy_fitness": 0.8125036559640642,
    "target_cdf": 0.9716911200959037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.191847801208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7075758306910231,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.191847801208496,
        "normalized": 0.9716911200959037,
        "contribution": 0.4858455600479519
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.71264910697937,
        "normalized": 0.8413447460685429,
        "contribution": 0.12620171191028143
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9295310974121094,
        "normalized": 0.6691256287429975,
        "contribution": 0.06691256287429975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5e7ac6decb75bedbdf1bd2e8b9744b8c770895eaeea8e60d1e12b9acfc7b1d6e",
    "id": "78d966a8ce99159b4fdd8885791a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 5.77639102935791,
    "genuineness": 4.218649387359619,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": 0.07847251743078232,
    "fitness": 0.7552785618393136,
    "legacy_fitness": 0.7552785618393136,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8044733238374295,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.387570858001709,
        "normalized": 0.36625184367446406,
        "contribution": 0.09156296091861602
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.997312545776367,
        "normalized": 0.8943502454552662,
        "contribution": 0.13415253681828992
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.862438201904297,
        "normalized": 0.897176290990003,
        "contribution": 0.0897176290990003
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d9e717b465bddbcd555d1257b275f1ccf3616f3c4685e0430205c8ce8af4ba5d",
    "id": "79c4579ec305e75a77f5a3c3f62b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": 0.48942479491233826,
    "voiceclap": -0.07315235584974289,
    "fitness": 0.7693962699472652,
    "legacy_fitness": 0.7693962699472652,
    "target_cdf": 0.8629676751891268,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.347759962081909,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6370478895589127,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.347759962081909,
        "normalized": 0.8629676751891268,
        "contribution": 0.4314838375945634
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.84308922290802,
        "normalized": 0.16633496146433524,
        "contribution": 0.04158374036608381
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1566660404205322,
        "normalized": 0.6955760628785511,
        "contribution": 0.10433640943178267
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7117059230804443,
        "normalized": 0.5964390216648275,
        "contribution": 0.05964390216648276
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fa57b09b4544e1207536e288b4d975509c48a30893c6d24106396f94518244f9",
    "id": "7a31f7b685f57bd413074d903665",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.607763767242432,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.19628658890724182,
    "fitness": 0.7955133119907672,
    "legacy_fitness": 0.7955133119907672,
    "target_cdf": 0.8824847618116329,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4451546669006348,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6049424681684428,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4451546669006348,
        "normalized": 0.8824847618116329,
        "contribution": 0.44124238090581647
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.58526611328125,
        "normalized": 0.10282367927847097,
        "contribution": 0.025705919819617742
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.952064275741577,
        "normalized": 0.6300671608645639,
        "contribution": 0.09451007412968458
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.2518529891967773,
        "normalized": 0.43484093313323924,
        "contribution": 0.043484093313323924
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a4b61011b5af81ee371da5d423998353d3c33ec8e2256470174cc7012f6d8d07",
    "id": "7a6a3b4b43cdc7874e6986e7f8b6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.117925643920898,
    "genuineness": 4.2751336097717285,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": 0.02150866761803627,
    "fitness": 0.7425633251459673,
    "legacy_fitness": 0.7425633251459673,
    "target_cdf": 0.929049708969098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7373390197753906,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7811645883816248,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7373390197753906,
        "normalized": 0.929049708969098,
        "contribution": 0.464524854484549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5656723976135254,
        "normalized": 0.44562653859681456,
        "contribution": 0.11140663464920364
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6548268795013428,
        "normalized": 0.828745318479122,
        "contribution": 0.1243117977718683
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4223875999450684,
        "normalized": 0.8092130147600393,
        "contribution": 0.08092130147600393
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "42c12eb0c5c86c8de7ab769317922c55f4c455ef2ce32abb9cfd18329aac5dd8",
    "id": "7a6c98cd4c713ffa506fadd01d3a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 3.6412534713745117,
    "wer": 0.0,
    "identity": 0.6477534770965576,
    "voiceclap": 0.11009367555379868,
    "fitness": 0.8056778813535032,
    "legacy_fitness": 0.8056778813535032,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8366107228570424,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.421494960784912,
        "normalized": 0.381046116164786,
        "contribution": 0.0952615290411965
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.135965347290039,
        "normalized": 0.9877755173761718,
        "contribution": 0.14816632760642576
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.328891754150391,
        "normalized": 0.9534909121221984,
        "contribution": 0.09534909121221985
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a008e753c4f603fb99cbf3e35e62c3166a979f250a25dbb007893bcdee2a6183",
    "id": "7aaafd34f1271ebfdcbecec8a8b5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.632867813110352,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03368864580988884,
    "fitness": 0.8269278090012216,
    "legacy_fitness": 0.8269278090012216,
    "target_cdf": 0.9511644047125467,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.932128429412842,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7011530421979368,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.932128429412842,
        "normalized": 0.9511644047125467,
        "contribution": 0.47558220235627335
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9041526317596436,
        "normalized": 0.18447617999078308,
        "contribution": 0.04611904499769577
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.588109016418457,
        "normalized": 0.8134348350041303,
        "contribution": 0.12201522525061953
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6478986740112305,
        "normalized": 0.574365695933482,
        "contribution": 0.0574365695933482
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0abb979ac51cffe59cb8748348611c016b542181d813327d5409697d85ec092d",
    "id": "7ad7bca1ac0c4725fe06990afa17",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07154898345470428,
    "fitness": 0.804839047437255,
    "legacy_fitness": 0.804839047437255,
    "target_cdf": 0.9796648197280255,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.337939739227295,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7796806582669327,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.337939739227295,
        "normalized": 0.9796648197280255,
        "contribution": 0.4898324098640128
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0602035522460938,
        "normalized": 0.23614752264221534,
        "contribution": 0.059036880660553834
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9152445793151855,
        "normalized": 0.9053242803940382,
        "contribution": 0.09053242803940383
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e968937536da0ce084af8dd9a03bb46a49b34a238f21eb0911bcb2716fa58968",
    "id": "7b5edc266bc1fb204a725223c636",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": 0.7193281054496765,
    "voiceclap": 0.0673343688249588,
    "fitness": 0.8106766475062273,
    "legacy_fitness": 0.8106766475062273,
    "target_cdf": 0.9922796672226393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7275185585021973,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7345330913543155,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7275185585021973,
        "normalized": 0.9922796672226393,
        "contribution": 0.49613983361131964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.3681519031524658,
        "normalized": 0.06480709428160003,
        "contribution": 0.016201773570400008
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.39762020111084,
        "normalized": 0.9453738037624609,
        "contribution": 0.14180607056436914
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4003851413726807,
        "normalized": 0.803854136082268,
        "contribution": 0.0803854136082268
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9be130ab2cda16a5e2a4568e098a203a5437683d2c89c6253a4157c72059070b",
    "id": "7b779b6b702f2a64ebf9299f29d2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": 0.5304677486419678,
    "voiceclap": -0.0899512767791748,
    "fitness": 0.8032770099280255,
    "legacy_fitness": 0.8032770099280255,
    "target_cdf": 0.912986084230339,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.623711585998535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7065962403513408,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.623711585998535,
        "normalized": 0.912986084230339,
        "contribution": 0.4564930421151695
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2637479305267334,
        "normalized": 0.3140598486863819,
        "contribution": 0.07851496217159548
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1322028636932373,
        "normalized": 0.6880159882453578,
        "contribution": 0.10320239823680367
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.975736379623413,
        "normalized": 0.6838583782777219,
        "contribution": 0.06838583782777219
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "14ca2ff4a1fd0ae6aa43bca62f53670232ddde6a8e87c31d2f12f15a7ebdb7fd",
    "id": "7bea2f5151b3ce9f2f4171471eb5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.8044304847717285,
    "wer": 0.0,
    "identity": 0.7181631326675415,
    "voiceclap": -0.004119252786040306,
    "fitness": 0.7442294895084609,
    "legacy_fitness": 0.7442294895084609,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6356454820249461,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3256850242614746,
        "normalized": 0.745423250776781,
        "contribution": 0.11181348761651715
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8085172176361084,
        "normalized": 0.6293295649240997,
        "contribution": 0.06293295649240997
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d1ab86ec2a5315b89e1a48a524ad910bd78d841a10faaa4217cea892c1d3dd88",
    "id": "7c49320d6c44332ff776fd5ef1b1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.229123115539551,
    "genuineness": 4.3127899169921875,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.04637251794338226,
    "fitness": 0.7569425361114742,
    "legacy_fitness": 0.7569425361114742,
    "target_cdf": 0.9511644047125467,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.932128429412842,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7227487583929443,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.932128429412842,
        "normalized": 0.9511644047125467,
        "contribution": 0.47558220235627335
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8804056644439697,
        "normalized": 0.1772798100559148,
        "contribution": 0.0443199525139787
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8193979263305664,
        "normalized": 0.8629676950479268,
        "contribution": 0.12944515425718903
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1407554149627686,
        "normalized": 0.7340144926550323,
        "contribution": 0.07340144926550322
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b482f1bbe4f9f2d1ef547bb84d11c42fd0de35f42ca27a081c66c9254852e540",
    "id": "7cbd5f1d330624f8b23313d0651f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 3.6192874908447266,
    "wer": 0.0,
    "identity": 0.6561466455459595,
    "voiceclap": 0.057948820292949677,
    "fitness": 0.7756122359952613,
    "legacy_fitness": 0.7756122359952613,
    "target_cdf": 0.9361753561737703,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7941524982452393,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7285120032846046,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7941524982452393,
        "normalized": 0.9361753561737703,
        "contribution": 0.4680876780868852
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0262794494628906,
        "normalized": 0.2242816522895154,
        "contribution": 0.05607041307237885
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.71264910697937,
        "normalized": 0.8413447460685429,
        "contribution": 0.12620171191028143
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.312375068664551,
        "normalized": 0.7815220021505915,
        "contribution": 0.07815220021505916
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "74ba07a216153a24a303c64d640c0391cba599a33a0ef1b8bd1d047a49cdb187",
    "id": "7ce6b43c24424a1992e0829f1ba8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.017816066741943,
    "wer": 0.0,
    "identity": 0.5584484934806824,
    "voiceclap": 0.04560322314500809,
    "fitness": 0.8159553068602138,
    "legacy_fitness": 0.8159553068602138,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7284874916303498,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6870384216308594,
        "normalized": 0.12539399986239474,
        "contribution": 0.031348499965598686
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9617295265197754,
        "normalized": 0.8885304554003988,
        "contribution": 0.1332795683100598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.175959348678589,
        "normalized": 0.7441683469516106,
        "contribution": 0.07441683469516107
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f2a6df437700add2628ac9dcda227bf6e6fe67e7ced875acefa07b75cd26b0be",
    "id": "7d1cd717a3e113922241c2cc758f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.2751336097717285,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.029866334050893784,
    "fitness": 0.8232271952467459,
    "legacy_fitness": 0.8232271952467459,
    "target_cdf": 0.9731772269879376,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2161965370178223,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7482267926993432,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2161965370178223,
        "normalized": 0.9731772269879376,
        "contribution": 0.4865886134939688
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.134836435317993,
        "normalized": 0.26342804068679404,
        "contribution": 0.06585701017169851
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4724645614624023,
        "normalized": 0.7849622943953769,
        "contribution": 0.11774434415930653
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3079745769500732,
        "normalized": 0.780368248743694,
        "contribution": 0.07803682487436941
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "83e109a1a8000eda9975a3a8af9f2ca2e796843628df15ab9d5ec9be998a03f3",
    "id": "7d4008872ebc7ff3632475fdbdd8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.005286898463964462,
    "fitness": 0.83597843101959,
    "legacy_fitness": 0.83597843101959,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7470993288091539,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.768456220626831,
        "normalized": 0.1457841285203917,
        "contribution": 0.03644603213009792
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.059582710266113,
        "normalized": 0.9040003875053659,
        "contribution": 0.1356000581258049
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2771711349487305,
        "normalized": 0.7721946355605077,
        "contribution": 0.07721946355605078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6d5c94879ead448d3bf179e11a27b7873d69aec1e6ec42101b186f8f17ed175e",
    "id": "7d48cd0a075cc0eb89151a1626e6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.557555198669434,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13935092091560364,
    "fitness": 0.8094662171639977,
    "legacy_fitness": 0.8094662171639977,
    "target_cdf": 0.912986084230339,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.623711585998535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6698982369359987,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.623711585998535,
        "normalized": 0.912986084230339,
        "contribution": 0.4564930421151695
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7548866271972656,
        "normalized": 0.14223933958928087,
        "contribution": 0.03555983489732022
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3879549503326416,
        "normalized": 0.7626471853600028,
        "contribution": 0.11439707780400041
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8239188194274902,
        "normalized": 0.634482821195086,
        "contribution": 0.0634482821195086
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2847d94ac4afbfd232d0109c551e8c4c27c1f6659169121cf99b675bf11df6b8",
    "id": "7d8b6a339c89f0ee89b91eb3a442",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": 0.7401476502418518,
    "voiceclap": -0.011640751734375954,
    "fitness": 0.8320916073765132,
    "legacy_fitness": 0.8320916073765132,
    "target_cdf": 0.996049248624995,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.971005439758301,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7667654359919072,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.971005439758301,
        "normalized": 0.996049248624995,
        "contribution": 0.4980246243124975
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.826127290725708,
        "normalized": 0.16150748713874752,
        "contribution": 0.04037687178468688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.3086628913879395,
        "normalized": 0.9361753524593768,
        "contribution": 0.1404263028689065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.756826162338257,
        "normalized": 0.8793763702581625,
        "contribution": 0.08793763702581625
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "11df2f3a938a6421bc824a10d02d5f5576d947a01f37e3c78bafb017342af91f",
    "id": "7d9983aa257691af7185384ebf07",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": 0.5804029107093811,
    "voiceclap": -0.04670635610818863,
    "fitness": 0.7864267394906698,
    "legacy_fitness": 0.7864267394906698,
    "target_cdf": 0.9390533101491965,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8185012340545654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7041519510423219,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8185012340545654,
        "normalized": 0.9390533101491965,
        "contribution": 0.46952665507459823
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.872772216796875,
        "normalized": 0.8729869940654369,
        "contribution": 0.13094804910981553
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1583573818206787,
        "normalized": 0.7391168284168952,
        "contribution": 0.07391168284168952
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "65897a883a2b47bd840cf7f55bef879dcac4eea92fbdbb08183b477a3602a431",
    "id": "7dfa2eab0e330bc6cff3386390b7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.7291181087493896,
    "wer": 0.0,
    "identity": 0.6747542023658752,
    "voiceclap": 0.0813048854470253,
    "fitness": 0.8108790556857662,
    "legacy_fitness": 0.8108790556857662,
    "target_cdf": 0.9971593102355156,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.084632873535156,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7973729411613126,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.084632873535156,
        "normalized": 0.9971593102355156,
        "contribution": 0.4985796551177578
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.043241500854492,
        "normalized": 0.23017181441662055,
        "contribution": 0.057542953604155136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.0825910568237305,
        "normalized": 0.9862070216770844,
        "contribution": 0.14793105325156267
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c72cef2fcdecdcd679074c8ecf86062826fcb6a4dacd88e51d4340103605b320",
    "id": "7e13121c5171493ea146f73d8035",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.532451152801514,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": -0.12312070280313492,
    "fitness": 0.7753735724241907,
    "legacy_fitness": 0.7753735724241907,
    "target_cdf": 0.9178304042951448,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6561765670776367,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6615855497457854,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6561765670776367,
        "normalized": 0.9178304042951448,
        "contribution": 0.4589152021475724
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1077394485473633,
        "normalized": 0.6803757957394871,
        "contribution": 0.10205636936092306
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6655006408691406,
        "normalized": 0.5804813217946941,
        "contribution": 0.058048132179469414
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "54b7833d050d0beb16f1de4b5047bcea06bdc96eb76cca2518ef14516ef832f0",
    "id": "7e50db8f2bbe9ac6a4a4fabc8be8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 3.7291181087493896,
    "wer": 0.0,
    "identity": 0.5784680843353271,
    "voiceclap": -0.06062387675046921,
    "fitness": 0.8013975774236215,
    "legacy_fitness": 0.8013975774236215,
    "target_cdf": 0.9690622995272662,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.151266574859619,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6908164184229519,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.151266574859619,
        "normalized": 0.9690622995272662,
        "contribution": 0.4845311497636331
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.32744300365448,
        "normalized": 0.05908511838886088,
        "contribution": 0.01477127959721522
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8816680908203125,
        "normalized": 0.8746060247794507,
        "contribution": 0.1311909037169176
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.731508255004883,
        "normalized": 0.6032308534518596,
        "contribution": 0.06032308534518596
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1855525f596303647482e9961d0c863c5e2687eb1732915d372f0dab49f0a8bb",
    "id": "7e7b187dd02333fa01987382a825",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.3964879512786865,
    "wer": 0.0,
    "identity": 0.7197300791740417,
    "voiceclap": 0.12346421182155609,
    "fitness": 0.7898411732774033,
    "legacy_fitness": 0.7898411732774033,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8153240971841484,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.338381052017212,
        "normalized": 0.34515387848215506,
        "contribution": 0.08628846962053877
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.593326568603516,
        "normalized": 0.9619219031491297,
        "contribution": 0.14428828547236947
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.24088191986084,
        "normalized": 0.9453738096886104,
        "contribution": 0.09453738096886105
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4d6b4565e4d2950c7d73914ada8cb1fab99a34846bfcf2f496a77cffefb15963",
    "id": "7f0728b3a93e29935052b1b1349a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.052178338170051575,
    "fitness": 0.8295655013088127,
    "legacy_fitness": 0.8295655013088127,
    "target_cdf": 0.9904545229601343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6463565826416016,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7790787219314012,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6463565826416016,
        "normalized": 0.9904545229601343,
        "contribution": 0.4952272614800671
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0805578231811523,
        "normalized": 0.24342971276978492,
        "contribution": 0.06085742819244623
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.086269855499268,
        "normalized": 0.9079316126330063,
        "contribution": 0.13618974189495092
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.695219039916992,
        "normalized": 0.8680429036393686,
        "contribution": 0.08680429036393687
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "aa65d54691f19b14bf73cdcd35c2695ef6d800505fc21d145f0281ac54cb4258",
    "id": "8050efc6ce8266a66c48fceeae18",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": 0.7003844976425171,
    "voiceclap": 0.06509648263454437,
    "fitness": 0.7982043175297027,
    "legacy_fitness": 0.7982043175297027,
    "target_cdf": 0.9371461924705993,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8022687435150146,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6860441575764176,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8022687435150146,
        "normalized": 0.9371461924705993,
        "contribution": 0.46857309623529964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5169429779052734,
        "normalized": 0.7962001195838251,
        "contribution": 0.11943001793757375
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8723244667053223,
        "normalized": 0.6505177836408061,
        "contribution": 0.06505177836408062
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7c1f2c57ed3ab164613ec9721aa48ca3aacfd79277c0a0f9d5bbaea3caf1b1a9",
    "id": "806b6ae5b5142aeac762bee8b3a6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.632867813110352,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07503535598516464,
    "fitness": 0.7914365233870776,
    "legacy_fitness": 0.7914365233870776,
    "target_cdf": 0.8762105190540217,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.412689685821533,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6498927137367213,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.412689685821533,
        "normalized": 0.8762105190540217,
        "contribution": 0.43810525952701085
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.917722225189209,
        "normalized": 0.18866878339787096,
        "contribution": 0.04716719584946774
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.005438804626465,
        "normalized": 0.6476213017017056,
        "contribution": 0.09714319525525585
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9471330642700195,
        "normalized": 0.674770631049869,
        "contribution": 0.0674770631049869
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "04688700023d3cf93ea18e5f905096828cd42118eb16a5771a00c34f2bb77b9b",
    "id": "80edefe062ad4e35ee148ab82491",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.080576419830322,
    "wer": 0.0,
    "identity": 0.506279706954956,
    "voiceclap": -0.022812308743596077,
    "fitness": 0.7409721300010088,
    "legacy_fitness": 0.7409721300010088,
    "target_cdf": 0.8217031250405714,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.169203042984009,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5993501939028851,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.169203042984009,
        "normalized": 0.8217031250405714,
        "contribution": 0.4108515625202857
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8498740196228027,
        "normalized": 0.16829178263767364,
        "contribution": 0.04207294565941841
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.9342727661132812,
        "normalized": 0.6241528877228036,
        "contribution": 0.09362293315842055
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.515883445739746,
        "normalized": 0.5280275256476042,
        "contribution": 0.05280275256476043
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0ef4659dd518053b6f3b54135f37150335026bc09a9f686fdd6f628ec4f0d4ac",
    "id": "813f3f55978e5c7315e7137f3e83",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.162164688110352,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.15818610787391663,
    "fitness": 0.7362525285984065,
    "legacy_fitness": 0.7362525285984065,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6503169934947838,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.223383903503418,
        "normalized": 0.7157670812548709,
        "contribution": 0.10736506218823062
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.270555257797241,
        "normalized": 0.44138410796643635,
        "contribution": 0.044138410796643636
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c766be913b01cd9d8fd73be8ef3769a4b9000d907b12def60dadd1fe604045f7",
    "id": "81819cbb3759f2b666964fc9bb62",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": 0.6839780211448669,
    "voiceclap": 0.02904808707535267,
    "fitness": 0.8097492727842047,
    "legacy_fitness": 0.8097492727842047,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7354865471152388,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5716965198516846,
        "normalized": 0.1000529371132764,
        "contribution": 0.0250132342783191
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.255288600921631,
        "normalized": 0.9301035297844544,
        "contribution": 0.13951552946766815
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3343775272369385,
        "normalized": 0.787238162630965,
        "contribution": 0.0787238162630965
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3790ff2d4e7866332e8d36972d1972543dc4617ed0888da16daa4e94e14919f4",
    "id": "8259865a8271bdfb41c7c4d1f522",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.53094482421875,
    "genuineness": 4.595211505889893,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.028212683275341988,
    "fitness": 0.8136559688317867,
    "legacy_fitness": 0.8136559688317867,
    "target_cdf": 0.9178304042951448,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6561765670776367,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6661557690932467,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6561765670776367,
        "normalized": 0.9178304042951448,
        "contribution": 0.4589152021475724
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3301327228546143,
        "normalized": 0.7466748313867635,
        "contribution": 0.11200122470801452
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.634697198867798,
        "normalized": 0.5697671225980729,
        "contribution": 0.0569767122598073
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ccb6168599a1b3e0a087a336e721556d718755cddd4933817d6f22a674989bbe",
    "id": "829d186f560c5880f3e4eaf26dda",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.538887023925781,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.030393723398447037,
    "fitness": 0.8229386793315198,
    "legacy_fitness": 0.8229386793315198,
    "target_cdf": 0.9862070255954634,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5002641677856445,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7501533990746401,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5002641677856445,
        "normalized": 0.9862070255954634,
        "contribution": 0.4931035127977317
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7548866271972656,
        "normalized": 0.14223933958928087,
        "contribution": 0.03555983489732022
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.532400369644165,
        "normalized": 0.8346379079796576,
        "contribution": 0.08346379079796577
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8aa3ceaab36b58fb40ea9f6d9602d7fa52e1fade9c8970d1c4a848759656af93",
    "id": "8308f47389790e19e6a41298da4f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.031152738258242607,
    "fitness": 0.8071849731142222,
    "legacy_fitness": 0.8071849731142222,
    "target_cdf": 0.962564179804901,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.06198787689209,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7245647514002523,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.06198787689209,
        "normalized": 0.962564179804901,
        "contribution": 0.4812820899024505
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8396968841552734,
        "normalized": 0.16536210020893177,
        "contribution": 0.04134052505223294
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8016064167022705,
        "normalized": 0.8595110162463788,
        "contribution": 0.1289266524369568
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.127553939819336,
        "normalized": 0.7301548400861202,
        "contribution": 0.07301548400861202
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1665e3756f47df669c5f61a1e433ba028262effd534815d31f0e34d5e8f1b89c",
    "id": "8320a2de88ec1ebe7854506303ad",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.570657730102539,
    "genuineness": 4.055472373962402,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.03362998738884926,
    "fitness": 0.8222846558358534,
    "legacy_fitness": 0.8222846558358534,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7421319120846085,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6870384216308594,
        "normalized": 0.12539399986239474,
        "contribution": 0.031348499965598686
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.326454162597656,
        "normalized": 0.9381054562342774,
        "contribution": 0.1407158184351416
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.426788330078125,
        "normalized": 0.8102739668234495,
        "contribution": 0.08102739668234496
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "95ee8fbfd0127213c3af5eba276e682897b9cb8f294a1f848b2fa8cfb2944ae5",
    "id": "838cc9f17966cf463a443a000d7d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 3.7040140628814697,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04659302532672882,
    "fitness": 0.808347133410397,
    "legacy_fitness": 0.808347133410397,
    "target_cdf": 0.9937903356798943,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.808681011199951,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7966205836119767,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.808681011199951,
        "normalized": 0.9937903356798943,
        "contribution": 0.49689516783994714
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0873427391052246,
        "normalized": 0.2458838571357379,
        "contribution": 0.061470964283934476
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.886885643005371,
        "normalized": 0.97888519494114,
        "contribution": 0.146832779241171
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.976851463317871,
        "normalized": 0.9142167224692412,
        "contribution": 0.09142167224692413
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4e21c9b6d26631343243d0d188422db97f1b6207fd00be0f3e25ba2d52b70718",
    "id": "83ac40407ff6b235b69077bdb070",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": 0.5829790830612183,
    "voiceclap": -0.025200573727488518,
    "fitness": 0.8212527625849653,
    "legacy_fitness": 0.8212527625849653,
    "target_cdf": 0.9811510787587977,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3704047203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7536487216865886,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3704047203063965,
        "normalized": 0.9811510787587977,
        "contribution": 0.49057553937939885
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9855704307556152,
        "normalized": 0.2105000137670302,
        "contribution": 0.05262500344175755
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.404785633087158,
        "normalized": 0.8049331523569412,
        "contribution": 0.08049331523569413
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ca2597493a842c39ab73ac337906fbbd575a0a422dc6435cbc9f7babd84f6339",
    "id": "83e570cbde07f0273b4aaf6b05e6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": 0.6974567174911499,
    "voiceclap": -0.04869164153933525,
    "fitness": 0.834445336964434,
    "legacy_fitness": 0.834445336964434,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7405147501481797,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8837981224060059,
        "normalized": 0.17829687836447006,
        "contribution": 0.044574219591117514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.039543628692627,
        "normalized": 0.7037290703249057,
        "contribution": 0.07037290703249058
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "43e9af03f58953f1f9f8df948bc90a20644361fee8376965535ed0f791822acc",
    "id": "840e426f15250f6f531583106cb0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 3,
    "generation": null,
    "ce": 6.538887023925781,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0349133275449276,
    "fitness": 0.8058865398185452,
    "legacy_fitness": 0.8058865398185452,
    "target_cdf": 0.947918719056646,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8996634483337402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7413962785204342,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8996634483337402,
        "normalized": 0.947918719056646,
        "contribution": 0.473959359528323
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1755452156066895,
        "normalized": 0.27895870121176836,
        "contribution": 0.06973967530294209
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5169429779052734,
        "normalized": 0.7962001195838251,
        "contribution": 0.11943001793757375
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3167755603790283,
        "normalized": 0.782672257515953,
        "contribution": 0.07826722575159531
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8e4d03b4471845f950b6fef2e2f348c4db3964227eb38502ef3090f87476981e",
    "id": "843644b2f1ed71f7d7459fba7d9c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.08537895977497101,
    "fitness": 0.8070870466563826,
    "legacy_fitness": 0.8070870466563826,
    "target_cdf": 0.9638222590155852,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0782203674316406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8068908857034862,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0782203674316406,
        "normalized": 0.9638222590155852,
        "contribution": 0.4819111295077926
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.479166030883789,
        "normalized": 0.4065890483509133,
        "contribution": 0.10164726208772833
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.012055397033691,
        "normalized": 0.91900911416671,
        "contribution": 0.091900911416671
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "65b4f28216ed277534c995e4250d58607d17e823b13e648d3aa15355c16595a1",
    "id": "848f96243084fc3d0b3bda64deb7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.660081624984741,
    "wer": 0.0,
    "identity": 0.7293370962142944,
    "voiceclap": -0.041997723281383514,
    "fitness": 0.7717788897503688,
    "legacy_fitness": 0.7717788897503688,
    "target_cdf": 0.9247120186784414,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.704874038696289,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7445862713190972,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.704874038696289,
        "normalized": 0.9247120186784414,
        "contribution": 0.4623560093392207
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0568110942840576,
        "normalized": 0.23494557063169752,
        "contribution": 0.05873639265792438
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9439380168914795,
        "normalized": 0.8855360174979171,
        "contribution": 0.13283040262468757
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9240455627441406,
        "normalized": 0.9066346669726454,
        "contribution": 0.09066346669726455
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "61e8ad5ea00e0236345bf5e9ba88b87d30e06c64ee83bd51bb4c64ec00a96c73",
    "id": "84fd7a79777ff418d7bd4fb2098b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.469690799713135,
    "wer": 0.0,
    "identity": 0.657768189907074,
    "voiceclap": -0.2587173879146576,
    "fitness": 0.7861860791530902,
    "legacy_fitness": 0.7861860791530902,
    "target_cdf": 0.8824847618116329,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4451546669006348,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6032793077719643,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4451546669006348,
        "normalized": 0.8824847618116329,
        "contribution": 0.44124238090581647
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6259750127792358,
        "normalized": 0.11146953771947371,
        "contribution": 0.027867384429868428
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8986899852752686,
        "normalized": 0.6122378010090498,
        "contribution": 0.09183567015135746
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.2188491821289062,
        "normalized": 0.42333872284921914,
        "contribution": 0.042333872284921914
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "233fedf95fdc3a25b4694949e0769c4366db5502c78a5aa6c304da9652b3a146",
    "id": "8567706deb1f3683bdc9ec33808f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 3.6914620399475098,
    "wer": 0.0,
    "identity": 0.676338255405426,
    "voiceclap": 0.07415267825126648,
    "fitness": 0.8035368211478837,
    "legacy_fitness": 0.8035368211478837,
    "target_cdf": 0.9891870513381258,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.597659111022949,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.793154387338176,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.597659111022949,
        "normalized": 0.9891870513381258,
        "contribution": 0.4945935256690629
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.063595771789551,
        "normalized": 0.23735276923124937,
        "contribution": 0.05933819230781234
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.762345314025879,
        "normalized": 0.9726892678230921,
        "contribution": 0.1459033901734638
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "562fab8bbe4e717f0e1cee3bca3e4caddfe79f09ce78f1a3dbb469d0d95b707b",
    "id": "859327a40bf1d38ff560f5c845c4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.634199142456055,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.01902191713452339,
    "fitness": 0.8252007923040282,
    "legacy_fitness": 0.8252007923040282,
    "target_cdf": 0.9650455080496254,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0944528579711914,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7052485599100243,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0944528579711914,
        "normalized": 0.9650455080496254,
        "contribution": 0.4825227540248127
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7209625244140625,
        "normalized": 0.13363422338747727,
        "contribution": 0.03340855584686932
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5792131423950195,
        "normalized": 0.8113311992293513,
        "contribution": 0.12169967988440268
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.951533555984497,
        "normalized": 0.6761757015393965,
        "contribution": 0.06761757015393965
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "567eb5137fe9607b8c7646509456c47d0d8b27accba54c781af245a6c691c67c",
    "id": "85aa52594104a42d06a65ffc371b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 3.798154592514038,
    "wer": 0.0,
    "identity": 0.6904395222663879,
    "voiceclap": 0.12597160041332245,
    "fitness": 0.8018062968163979,
    "legacy_fitness": 0.8018062968163979,
    "target_cdf": 0.9736578860102679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2243127822875977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.777172874997157,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2243127822875977,
        "normalized": 0.9736578860102679,
        "contribution": 0.48682894300513396
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0805578231811523,
        "normalized": 0.24342971276978492,
        "contribution": 0.06085742819244623
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.424307346343994,
        "normalized": 0.9479187150078625,
        "contribution": 0.14218780725117938
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7216219902038574,
        "normalized": 0.8729869654839748,
        "contribution": 0.08729869654839749
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2884cb0e6442f4e971185f31c18c270a3c49c81b315d1fe0dd39d82f89594944",
    "id": "860e1c44a9b95c4a3d7a6a2c3f6b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": 0.46867167949676514,
    "voiceclap": -0.03583849221467972,
    "fitness": 0.8202088393891698,
    "legacy_fitness": 0.8202088393891698,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8225297061175851,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3621277809143066,
        "normalized": 0.3552838131596231,
        "contribution": 0.08882095328990577
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.388724327087402,
        "normalized": 0.9445039636441842,
        "contribution": 0.14167559454662762
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.27608585357666,
        "normalized": 0.9487457744888368,
        "contribution": 0.09487457744888368
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0c0a572c21c3f526704845b7c8c4c4dc0a6176c4a1afc3996e25e2989b6b8f31",
    "id": "863fa0a6391e9940ba9951ac2072",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 3.948779582977295,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.016726745292544365,
    "fitness": 0.8200515195742715,
    "legacy_fitness": 0.8200515195742715,
    "target_cdf": 0.9872707933218218,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.532729148864746,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7350142240339453,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.532729148864746,
        "normalized": 0.9872707933218218,
        "contribution": 0.4936353966609109
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5784813165664673,
        "normalized": 0.10143141696818753,
        "contribution": 0.02535785424204688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.255288600921631,
        "normalized": 0.9301035297844544,
        "contribution": 0.13951552946766815
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.250767946243286,
        "normalized": 0.765054436633194,
        "contribution": 0.0765054436633194
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7bc5d5c1c529208ace6c9e71420f84b589ced60ae8cf79355289ece88be2b1a8",
    "id": "8640147984de5257acab830fa238",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": 0.4531276524066925,
    "voiceclap": -0.01956581510603428,
    "fitness": 0.7914605067147985,
    "legacy_fitness": 0.7914605067147985,
    "target_cdf": 0.9351928961225698,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.786036252975464,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6899363520292389,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.786036252975464,
        "normalized": 0.9351928961225698,
        "contribution": 0.4675964480612849
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.570317506790161,
        "normalized": 0.8092130613886472,
        "contribution": 0.12138195920829709
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9229302406311035,
        "normalized": 0.6669986716810576,
        "contribution": 0.06669986716810576
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bac6f81c0c19e44f76bdfe31b220803b6fbc083717fdfb219503e7e6e8a1e01d",
    "id": "8699e2a6259f45a56d0a3f1c24db",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.010832792147994041,
    "fitness": 0.8357675251798591,
    "legacy_fitness": 0.8357675251798591,
    "target_cdf": 0.9981710548076552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.230724811553955,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7616327201384567,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.230724811553955,
        "normalized": 0.9981710548076552,
        "contribution": 0.4990855274038276
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.632759928703308,
        "normalized": 0.11295969157872005,
        "contribution": 0.02823992289468001
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.744553565979004,
        "normalized": 0.9716911067853843,
        "contribution": 0.14575366601780765
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7920303344726562,
        "normalized": 0.8855360382214144,
        "contribution": 0.08855360382214145
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b4d656c8d4da9e04dc943a1d0c31d9bb10c2d171b9a8ca967872756bd243bc15",
    "id": "87561d3ed1aa3be1ae65722da90b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.1998209953308105,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06258435547351837,
    "fitness": 0.8240350360566487,
    "legacy_fitness": 0.8240350360566487,
    "target_cdf": 0.9755090862302205,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.256777286529541,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6757227381637462,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.256777286529541,
        "normalized": 0.9755090862302205,
        "contribution": 0.48775454311511024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.490278720855713,
        "normalized": 0.08456572900066728,
        "contribution": 0.02114143225016682
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4012985229492188,
        "normalized": 0.7662529752285923,
        "contribution": 0.11493794628428883
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.490030527114868,
        "normalized": 0.518888165141803,
        "contribution": 0.05188881651418031
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8cb134ba698751abf9f735939eaca8214d42e5419f24fdeb4af58492aa2bcc0d",
    "id": "87bc5efe03455f105deddcc97599",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 3.672633647918701,
    "wer": 0.0,
    "identity": 0.6309126615524292,
    "voiceclap": 0.09552793204784393,
    "fitness": 0.8117160691050893,
    "legacy_fitness": 0.8117160691050893,
    "target_cdf": 0.9999661710318338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.350764274597168,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8273383150524911,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.350764274597168,
        "normalized": 0.9999661710318338,
        "contribution": 0.4999830855159169
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2756214141845703,
        "normalized": 0.3189263670056157,
        "contribution": 0.07973159175140393
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.669709205627441,
        "normalized": 0.9967235423007876,
        "contribution": 0.14950853134511813
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.777743339538574,
        "normalized": 0.981151064400521,
        "contribution": 0.09811510644005211
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "01d754941f2018aa645bd38293a5bc6dea4252f561561ad70c5a6a11bd658b06",
    "id": "87eba4b7608b20669fb3e5743c9c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.3127899169921875,
    "wer": 0.0,
    "identity": 0.7031475901603699,
    "voiceclap": -0.02762971818447113,
    "fitness": 0.8265819475665592,
    "legacy_fitness": 0.8265819475665592,
    "target_cdf": 0.9759544383714404,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2648935317993164,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7383464156385121,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2648935317993164,
        "normalized": 0.9759544383714404,
        "contribution": 0.4879772191857202
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.917722225189209,
        "normalized": 0.18866878339787096,
        "contribution": 0.04716719584946774
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.694857597351074,
        "normalized": 0.8375344060053678,
        "contribution": 0.12563016090080517
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.290372610092163,
        "normalized": 0.7757183970251897,
        "contribution": 0.07757183970251898
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "42f46268023874f579440947f4e33c2a9838a9d279ce36965d9303ac0c7bb20a",
    "id": "87ff379748ef689c3efa2c81008c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.061748504638672,
    "wer": 0.0,
    "identity": 0.8061049580574036,
    "voiceclap": -0.025991465896368027,
    "fitness": 0.7545244469292567,
    "legacy_fitness": 0.7545244469292567,
    "target_cdf": 0.8487882165208668,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.282830238342285,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6260869527229996,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.282830238342285,
        "normalized": 0.8487882165208668,
        "contribution": 0.4243941082604334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3568201065063477,
        "normalized": 0.7541161908392326,
        "contribution": 0.11311742862588489
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6875033378601074,
        "normalized": 0.5880985182046271,
        "contribution": 0.058809851820462715
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "89bfe556294c05b7994e5ab6fb71aba981f2dfd3d6fe8868231d56fd74564f83",
    "id": "88330d5d2a72de6920ff3301147c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": 0.6011927723884583,
    "voiceclap": -0.15975958108901978,
    "fitness": 0.8055010639881435,
    "legacy_fitness": 0.8055010639881435,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6765287909251954,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9007601737976074,
        "normalized": 0.18343713514349452,
        "contribution": 0.04585928378587363
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.21004056930542,
        "normalized": 0.7117805375367806,
        "contribution": 0.10676708063051708
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8349201679229736,
        "normalized": 0.6381490431448433,
        "contribution": 0.06381490431448433
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "541eb9b336c67ac3085bdaee130ebcbe9b85fbc2f5946220075cd468188bad44",
    "id": "88577799523da29c6c5f1ca87fbb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.197352409362793,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": 0.7420566082000732,
    "voiceclap": -0.08950091898441315,
    "fitness": 0.8202790262437707,
    "legacy_fitness": 0.8202790262437707,
    "target_cdf": 0.9592621807529222,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.021406888961792,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.709958248993609,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.021406888961792,
        "normalized": 0.9592621807529222,
        "contribution": 0.4796310903764611
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9686083793640137,
        "normalized": 0.20490763245182997,
        "contribution": 0.05122690811295749
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4190900325775146,
        "normalized": 0.7710131225645452,
        "contribution": 0.11565196838468178
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8239188194274902,
        "normalized": 0.634482821195086,
        "contribution": 0.0634482821195086
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8b8642573b72a83436d371c42af57f31a84cb0023642d7a2b3add43c5835b384",
    "id": "88923615662079235829d613a80f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 4.406930446624756,
    "wer": 0.0,
    "identity": 0.41202807426452637,
    "voiceclap": -0.10334708541631699,
    "fitness": 0.8357688427928824,
    "legacy_fitness": 0.8357688427928824,
    "target_cdf": 0.9904545229601343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6463565826416016,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6967483687980056,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6463565826416016,
        "normalized": 0.9904545229601343,
        "contribution": 0.4952272614800671
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.2867341041564941,
        "normalized": 0.05376718151292953,
        "contribution": 0.013441795378232382
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.667701005935669,
        "normalized": 0.5812444830996799,
        "contribution": 0.05812444830996799
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b76205f051f0947dc8bff891e7921228c4579be57a0acdec0a8df3729cad64fa",
    "id": "891ce6d786ba00fa0a4f2c5e0aee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.515059471130371,
    "genuineness": 4.074300289154053,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03968600928783417,
    "fitness": 0.8151617528511576,
    "legacy_fitness": 0.8151617528511576,
    "target_cdf": 0.9650455080496254,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0944528579711914,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7480306179760271,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0944528579711914,
        "normalized": 0.9650455080496254,
        "contribution": 0.4825227540248127
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.073773145675659,
        "normalized": 0.24098894914960267,
        "contribution": 0.06024723728740067
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.730440616607666,
        "normalized": 0.845096013305189,
        "contribution": 0.12676440199577835
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3255765438079834,
        "normalized": 0.7849622466803536,
        "contribution": 0.07849622466803537
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f21bd0d242a4968f857add4a06d400f7f5974f9ff9f07dd1a2dfc8b647849878",
    "id": "89662db938aaf2f4c0b1b44f2fc8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1934562474489212,
    "fitness": 0.8019527299863353,
    "legacy_fitness": 0.8019527299863353,
    "target_cdf": 0.9258149432464946,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7129902839660645,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.730053369662721,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7129902839660645,
        "normalized": 0.9258149432464946,
        "contribution": 0.4629074716232473
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.097519874572754,
        "normalized": 0.24958974345161927,
        "contribution": 0.06239743586290482
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.570317506790161,
        "normalized": 0.8092130613886472,
        "contribution": 0.12138195920829709
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5279998779296875,
        "normalized": 0.8336650296827179,
        "contribution": 0.0833665029682718
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "affd73791c7c2f3af14010bf8989aeb1ff4513eeb197527728b3a6706d683cbc",
    "id": "8974ddfadd4cc085df9348c020be",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04904066026210785,
    "fitness": 0.8283351694438965,
    "legacy_fitness": 0.8283351694438965,
    "target_cdf": 0.9755090862302205,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.256777286529541,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6983804669275058,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.256777286529541,
        "normalized": 0.9755090862302205,
        "contribution": 0.48775454311511024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.415645718574524,
        "normalized": 0.07201630336001663,
        "contribution": 0.018004075840004158
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.068478107452393,
        "normalized": 0.9053242349151492,
        "contribution": 0.13579863523727237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6302967071533203,
        "normalized": 0.56823212735119,
        "contribution": 0.056823212735119005
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "472093faeddd031871c9f7fa6aeb45f5c37709ebef82fb778dfd3f078c247c1d",
    "id": "8a662e147e9ada04470ed15d57f2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": 0.5603185296058655,
    "voiceclap": -0.06618554145097733,
    "fitness": 0.8372829707295505,
    "legacy_fitness": 0.8372829707295505,
    "target_cdf": 0.9935110473296658,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7924485206604004,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7400829781681864,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7924485206604004,
        "normalized": 0.9935110473296658,
        "contribution": 0.4967555236648329
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.997312545776367,
        "normalized": 0.8943502454552662,
        "contribution": 0.13415253681828992
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.193561553955078,
        "normalized": 0.7491684009351236,
        "contribution": 0.07491684009351236
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ed10208a2f296c3c98e63f16897f4ff80e37c520ead390fefe75f99cbfab87ca",
    "id": "8a8150242f182ab75ab1e0d36e6c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.337893962860107,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07252401113510132,
    "fitness": 0.8400352269021685,
    "legacy_fitness": 0.8400352269021685,
    "target_cdf": 0.9964005449628859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.003470420837402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7885468068547198,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.003470420837402,
        "normalized": 0.9964005449628859,
        "contribution": 0.49820027248144294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0228869915008545,
        "normalized": 0.22311395750311608,
        "contribution": 0.05577848937577902
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.620013236999512,
        "normalized": 0.9638222528181529,
        "contribution": 0.14457333792272292
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.880040407180786,
        "normalized": 0.899947070747749,
        "contribution": 0.08999470707477492
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cf211700dd9576aef023d9d41293d1fe9d6b16320f608fcf95a60667247fa5a0",
    "id": "8adde953ff6841b135685032f94f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13734856247901917,
    "fitness": 0.7750010069684486,
    "legacy_fitness": 0.7750010069684486,
    "target_cdf": 0.8885304563197098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4776196479797363,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.643424388873171,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4776196479797363,
        "normalized": 0.8885304563197098,
        "contribution": 0.4442652281598549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.223383903503418,
        "normalized": 0.7157670812548709,
        "contribution": 0.10736506218823062
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7931153774261475,
        "normalized": 0.6241529208590221,
        "contribution": 0.06241529208590221
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "39ebbaa45510dbe3ba1f7c49f100bdba0b5b52ebf1563e2850c780fb61b25522",
    "id": "8bb33e262565fc563aa05671d778",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.086852550506592,
    "wer": 0.0,
    "identity": 0.5591666102409363,
    "voiceclap": -0.10952334851026535,
    "fitness": 0.7524444024341432,
    "legacy_fitness": 0.7524444024341432,
    "target_cdf": 0.8413447460685429,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2503652572631836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.621135784522085,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2503652572631836,
        "normalized": 0.8413447460685429,
        "contribution": 0.42067237303427146
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.707392930984497,
        "normalized": 0.130294529860791,
        "contribution": 0.03257363246519775
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3390283584594727,
        "normalized": 0.7491683218823387,
        "contribution": 0.1123752482823508
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5928924083709717,
        "normalized": 0.5551453074026497,
        "contribution": 0.05551453074026497
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "41f683a46e055848e127f03bde1f6036cc11c2f714035a05eac7d394ce843b58",
    "id": "8bdd88b000a6ab9a4ac87767378e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.645420074462891,
    "wer": 0.07142857142857142,
    "identity": null,
    "voiceclap": -0.12744726240634918,
    "fitness": 0.7721562263524447,
    "legacy_fitness": 0.7721562263524447,
    "target_cdf": 0.9585742155738795,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0132906436920166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6667558500234585,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0132906436920166,
        "normalized": 0.9585742155738795,
        "contribution": 0.4792871077869397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4834939241409302,
        "normalized": 0.0833611930352981,
        "contribution": 0.020840298258824524
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.352372169494629,
        "normalized": 0.7528841549820324,
        "contribution": 0.11293262324730485
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5411863327026367,
        "normalized": 0.5369582073038938,
        "contribution": 0.05369582073038938
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "aad56df26c827ff97a156762ba176b4423c4fc32b01aea913a4761f2743f50a4",
    "id": "8bf9547ea09c878ea83126f8aa66",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.936227321624756,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.1696094423532486,
    "fitness": 0.806380934271703,
    "legacy_fitness": 0.806380934271703,
    "target_cdf": 0.962564179804901,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.06198787689209,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7518576255761988,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.06198787689209,
        "normalized": 0.962564179804901,
        "contribution": 0.4812820899024505
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1110897064208984,
        "normalized": 0.25457682692833283,
        "contribution": 0.06364420673208321
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.792710781097412,
        "normalized": 0.8577606879337987,
        "contribution": 0.1286641031900698
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3167755603790283,
        "normalized": 0.782672257515953,
        "contribution": 0.07826722575159531
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7488dad248373982a80a86694357ac87c64000dc78067d1d7b43176c6697f771",
    "id": "8c785c288669733cee246f3220b0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.197352409362793,
    "genuineness": 4.231201171875,
    "wer": 0.06666666666666667,
    "identity": 0.2023146152496338,
    "voiceclap": -0.12919291853904724,
    "fitness": 0.7544665822747494,
    "legacy_fitness": 0.7544665822747494,
    "target_cdf": 0.9542463867500731,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9645934104919434,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7497827926081507,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9645934104919434,
        "normalized": 0.9542463867500731,
        "contribution": 0.47712319337503656
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.063595771789551,
        "normalized": 0.23735276923124937,
        "contribution": 0.05933819230781234
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.872772216796875,
        "normalized": 0.8729869940654369,
        "contribution": 0.13094804910981553
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.483994960784912,
        "normalized": 0.8237335781548634,
        "contribution": 0.08237335781548634
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1010318df8cf07b21bd786627b85eee06388b8fdca7ecf563e0d79ac8dde137b",
    "id": "8ca244fcb158b683c7d567e0ae87",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.19953879714012146,
    "fitness": 0.826354933283874,
    "legacy_fitness": 0.826354933283874,
    "target_cdf": 0.9619218677325021,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0538716316223145,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6960853732360379,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0538716316223145,
        "normalized": 0.9619218677325021,
        "contribution": 0.48096093386625105
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.605900287628174,
        "normalized": 0.8175982108486767,
        "contribution": 0.12263973162730149
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.731508255004883,
        "normalized": 0.6032308534518596,
        "contribution": 0.06032308534518596
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f6a05c37e1d7c2b4d96040c0f29e37129f1589bc5ab1ed9dac3ffd247d9707d2",
    "id": "8ce864691dffa5fb1148e8e698a0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": 0.6377732157707214,
    "voiceclap": -0.05391480773687363,
    "fitness": 0.8202260540663983,
    "legacy_fitness": 0.8202260540663983,
    "target_cdf": 0.9862070255954634,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5002641677856445,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7586619411598732,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5002641677856445,
        "normalized": 0.9862070255954634,
        "contribution": 0.4931035127977317
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9041526317596436,
        "normalized": 0.18447617999078308,
        "contribution": 0.04611904499769577
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8816680908203125,
        "normalized": 0.8746060247794507,
        "contribution": 0.1311909037169176
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.774428367614746,
        "normalized": 0.8824847964752807,
        "contribution": 0.08824847964752808
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0af9d66ad6cec846cc86553db3163cb99b733b5b48f140ba905484b7b2ce9967",
    "id": "8d05a932e1ac8ffd356465388931",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09545916318893433,
    "fitness": 0.7680711644156363,
    "legacy_fitness": 0.7680711644156363,
    "target_cdf": 0.8840175146388185,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.45327091217041,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.629181251688534,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.45327091217041,
        "normalized": 0.8840175146388185,
        "contribution": 0.44200875731940925
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.285654067993164,
        "normalized": 0.7340144537731833,
        "contribution": 0.11010216806597749
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.560438632965088,
        "normalized": 0.5437411126781538,
        "contribution": 0.054374111267815384
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f57fd7ed9b0c89f879831a69c480329b7f70f3e12ae281279c1da1b7f250fccc",
    "id": "8d08c50a3bbce2001c17097e5bf9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.545003414154053,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.053589385002851486,
    "fitness": 0.8484208746427564,
    "legacy_fitness": 0.8484208746427564,
    "target_cdf": 0.9919413812308048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7112860679626465,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7302220216254283,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7112860679626465,
        "normalized": 0.9919413812308048,
        "contribution": 0.4959706906154024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.021941661834717,
        "normalized": 0.6983048327926806,
        "contribution": 0.06983048327926807
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7c35d7403e3c6b717a9ffaa7387f1523050fec4ba423dba50eb12086c27dbe1a",
    "id": "8d3d41ce8ff8aeb0e25f13f0f0ab",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.607763767242432,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.16288310289382935,
    "fitness": 0.763777616034533,
    "legacy_fitness": 0.763777616034533,
    "target_cdf": 0.8277502902559148,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.193551540374756,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.610733407145157,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.193551540374756,
        "normalized": 0.8277502902559148,
        "contribution": 0.4138751451279574
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0161020755767822,
        "normalized": 0.22078898327927737,
        "contribution": 0.05519724581981934
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.89424204826355,
        "normalized": 0.6107406663737006,
        "contribution": 0.09161109995605508
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4380838871002197,
        "normalized": 0.500499162413252,
        "contribution": 0.0500499162413252
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b153001eb909629058e9a61e6aad648a6b5660bdc96a2634cd223a8f66b758cc",
    "id": "8d45d72f3e0daebc9c84becefd99",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 5.967014789581299,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07841628789901733,
    "fitness": 0.818418006031818,
    "legacy_fitness": 0.818418006031818,
    "target_cdf": 0.9994832593780891,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.620303630828857,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7865785317338152,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.620303630828857,
        "normalized": 0.9994832593780891,
        "contribution": 0.49974162968904456
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9007601737976074,
        "normalized": 0.18343713514349452,
        "contribution": 0.04585928378587363
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.047008514404297,
        "normalized": 0.9850674568917444,
        "contribution": 0.14776011853376164
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.1176676750183105,
        "normalized": 0.9321749972513534,
        "contribution": 0.09321749972513535
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f47826afa11eccf84ab20d98b63d23aca8bce6bd7dcb177bd431708a3d1e0d00",
    "id": "8d66127f68d58b43ac3f76e19201",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": 0.7043946385383606,
    "voiceclap": 0.030407065525650978,
    "fitness": 0.8199515881279384,
    "legacy_fitness": 0.8199515881279384,
    "target_cdf": 0.9716911200959037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.191847801208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7289244874315793,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.191847801208496,
        "normalized": 0.9716911200959037,
        "contribution": 0.4858455600479519
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.068478107452393,
        "normalized": 0.9053242349151492,
        "contribution": 0.13579863523727237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1715588569641113,
        "normalized": 0.7429102710689157,
        "contribution": 0.07429102710689157
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1b6b7b4b1815ea138dcf7924e3c17edf3e986b0e07b634465acda309a3c5026d",
    "id": "8d70f8abfbc17b82c335890877b5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 5.982900142669678,
    "genuineness": 4.118232727050781,
    "wer": 0.06666666666666667,
    "identity": 0.5503552556037903,
    "voiceclap": -0.0405782125890255,
    "fitness": 0.7514671421320022,
    "legacy_fitness": 0.7514671421320022,
    "target_cdf": 0.9679556654189556,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1350340843200684,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7792645350564391,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1350340843200684,
        "normalized": 0.9679556654189556,
        "contribution": 0.4839778327094778
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2400012016296387,
        "normalized": 0.3044239560639862,
        "contribution": 0.07610598901599655
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.952833652496338,
        "normalized": 0.8870402984992682,
        "contribution": 0.13305604477489022
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.660015106201172,
        "normalized": 0.861246685560745,
        "contribution": 0.08612466855607451
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "eaf20ecf5d923916cde64c5d5c537855b9f2a079005de241815b480f3ff77aba",
    "id": "8f2bc746f146f8abf73a9405e4f4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.319066047668457,
    "wer": 0.03125,
    "identity": null,
    "voiceclap": 0.008381111547350883,
    "fitness": 0.7940926707171126,
    "legacy_fitness": 0.7940926707171126,
    "target_cdf": 0.9557276754241297,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.980825901031494,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7301302784519677,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.980825901031494,
        "normalized": 0.9557276754241297,
        "contribution": 0.47786383771206487
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8600512742996216,
        "normalized": 0.17125467685227086,
        "contribution": 0.042813669213067715
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.0239996910095215,
        "normalized": 0.8985685934084969,
        "contribution": 0.13478528901127454
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.184760332107544,
        "normalized": 0.7466748251556059,
        "contribution": 0.07466748251556059
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4ff14ba36878c275c903152f9bf6d8385dccec0debd208c80c8dc7da03ea837b",
    "id": "8f5d7e3ced5ae60f03f5267495bd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": 0.45022839307785034,
    "voiceclap": 0.09706425666809082,
    "fitness": 0.8398286919066982,
    "legacy_fitness": 0.8398286919066982,
    "target_cdf": 0.9971593102355156,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.084632873535156,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8021772162753826,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.084632873535156,
        "normalized": 0.9971593102355156,
        "contribution": 0.4985796551177578
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2026846408843994,
        "normalized": 0.28955421544612325,
        "contribution": 0.07238855386153081
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.415411949157715,
        "normalized": 0.9470811189066717,
        "contribution": 0.14206216783600076
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8272342681884766,
        "normalized": 0.891468394600932,
        "contribution": 0.0891468394600932
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4f82588f9e622cb8d0bbe5ac32369c74ce81939a34c0fbe487f322492e2df8b0",
    "id": "901d536d41f39c3e5cf7a8e4cde5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.538887023925781,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.05662619695067406,
    "fitness": 0.7937464075454362,
    "legacy_fitness": 0.7937464075454362,
    "target_cdf": 0.8943502420521354,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.510084629058838,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6500344820582763,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.510084629058838,
        "normalized": 0.8943502420521354,
        "contribution": 0.4471751210260677
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.241175413131714,
        "normalized": 0.7210412211477328,
        "contribution": 0.10815618317215993
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6060938835144043,
        "normalized": 0.5597717954445941,
        "contribution": 0.05597717954445941
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1c6f8ae2649081b1b08b69a509c1a362a80a84440dcffd3d0a91fe74dd330c17",
    "id": "9020689f3975b4d3a384b13795fb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": 0.5834708213806152,
    "voiceclap": 0.09553875029087067,
    "fitness": 0.8116611995520555,
    "legacy_fitness": 0.8116611995520555,
    "target_cdf": 0.9768247481628979,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.281126022338867,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7348187369510282,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.281126022338867,
        "normalized": 0.9768247481628979,
        "contribution": 0.48841237408144894
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.015103816986084,
        "normalized": 0.8971763038367273,
        "contribution": 0.1345764455755091
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.290372610092163,
        "normalized": 0.7757183970251897,
        "contribution": 0.07757183970251898
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bb71fd28253eef6b693355085cf3dd05e7a0ec3b389cc7b9dc4e96216739a497",
    "id": "903b8396104a5ca128bab70f89e2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.051640965044498444,
    "fitness": 0.8215239025874985,
    "legacy_fitness": 0.8215239025874985,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.738901238066474,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8994593620300293,
        "normalized": 0.8778006492473432,
        "contribution": 0.13167009738710148
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.347579002380371,
        "normalized": 0.790625537042531,
        "contribution": 0.07906255370425311
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d42a691a733a5c9f3278768bb86fe0601ffa8cf57ee1f0db4a3a737a8848592c",
    "id": "906ba55f12e6c08b551c59d9abfa",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": 0.6809917688369751,
    "voiceclap": -0.035392340272665024,
    "fitness": 0.8403728184348119,
    "legacy_fitness": 0.8403728184348119,
    "target_cdf": 0.996049248624995,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.971005439758301,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.743775222297558,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.971005439758301,
        "normalized": 0.996049248624995,
        "contribution": 0.4980246243124975
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7209625244140625,
        "normalized": 0.13363422338747727,
        "contribution": 0.03340855584686932
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.184760332107544,
        "normalized": 0.7466748251556059,
        "contribution": 0.07466748251556059
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3ac8b671976591ccc7ced846cdaade09b832e9856f99e349c6be6a3b53b5ee75",
    "id": "909f12724d344228f8a3e6b6061c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.873466968536377,
    "wer": 0.0,
    "identity": 0.710731029510498,
    "voiceclap": -0.03633035719394684,
    "fitness": 0.774861104540587,
    "legacy_fitness": 0.774861104540587,
    "target_cdf": 0.9079315965425183,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5912468433380127,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.700545899645021,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5912468433380127,
        "normalized": 0.9079315965425183,
        "contribution": 0.45396579827125916
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.84308922290802,
        "normalized": 0.16633496146433524,
        "contribution": 0.04158374036608381
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.774919033050537,
        "normalized": 0.8542158295589685,
        "contribution": 0.12813237443384526
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2639694213867188,
        "normalized": 0.7686398657383282,
        "contribution": 0.07686398657383282
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9a4ea8ac967ff0085f27f3dce4193ffab19794f68db47c501ed19d7bc3ba2b79",
    "id": "90bb0333c4d53f6ec480c36520c7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.109982967376709,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": 0.6457370519638062,
    "voiceclap": 0.09028155356645584,
    "fitness": 0.8206425572834531,
    "legacy_fitness": 0.8206425572834531,
    "target_cdf": 0.9999219785932938,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.139742374420166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7875340201226058,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.139742374420166,
        "normalized": 0.9999219785932938,
        "contribution": 0.4999609892966469
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.367254734039307,
        "normalized": 0.9929189456061928,
        "contribution": 0.1489378418409289
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.355295181274414,
        "normalized": 0.9557276727419275,
        "contribution": 0.09557276727419275
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0c0f87656d1da596da49259fc35753d5e872e3fd743cfaa68d9d1c08cf8a3643",
    "id": "919a0bff7c0b272df66dadbeb320",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.469690799713135,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13241085410118103,
    "fitness": 0.8258286543225148,
    "legacy_fitness": 0.8258286543225148,
    "target_cdf": 0.962564179804901,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.06198787689209,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7248814918022967,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.06198787689209,
        "normalized": 0.962564179804901,
        "contribution": 0.4812820899024505
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8939753770828247,
        "normalized": 0.18137005253541438,
        "contribution": 0.045342513133853596
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.694857597351074,
        "normalized": 0.8375344060053678,
        "contribution": 0.12563016090080517
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.114352226257324,
        "normalized": 0.7262672786518737,
        "contribution": 0.07262672786518737
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "77d18b41c7730f26a5463e98a52f6b833f359e92b42305ef046915a0ac2bfe61",
    "id": "9242f9e9d02c9adbc674fa872d9a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.594485759735107,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.018987631425261497,
    "fitness": 0.8009800287717093,
    "legacy_fitness": 0.8009800287717093,
    "target_cdf": 0.929049708969098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7373390197753906,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6798218308689513,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7373390197753906,
        "normalized": 0.929049708969098,
        "contribution": 0.464524854484549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.707392930984497,
        "normalized": 0.130294529860791,
        "contribution": 0.03257363246519775
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4991517066955566,
        "normalized": 0.7917476292303571,
        "contribution": 0.11876214438455357
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.839320659637451,
        "normalized": 0.6396119953465103,
        "contribution": 0.06396119953465103
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d8c6fc1c432149e2963be1050d2245d5644f929f09c56b061bbb5900298ece2c",
    "id": "92494bbeac7459cf6d8d00f3b96e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 3.5188708305358887,
    "wer": 0.0,
    "identity": 0.701409637928009,
    "voiceclap": 0.10535812377929688,
    "fitness": 0.795568315843175,
    "legacy_fitness": 0.795568315843175,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7908852853638315,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0771656036376953,
        "normalized": 0.24220770725406027,
        "contribution": 0.06055192681351507
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.584430694580078,
        "normalized": 0.9612706211599873,
        "contribution": 0.1441905931739981
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d3ce013ef05c78ef0a981f047ef835de97b20cac905520f2187f7c1363e66e15",
    "id": "927ce2dd4390dbee63d629ff77ae",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": 0.6594431400299072,
    "voiceclap": -0.02063898928463459,
    "fitness": 0.8348439503327192,
    "legacy_fitness": 0.8348439503327192,
    "target_cdf": 0.9989584622015916,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.4092817306518555,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.74988539359498,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.4092817306518555,
        "normalized": 0.9989584622015916,
        "contribution": 0.4994792311007958
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6463295221328735,
        "normalized": 0.11598248129215816,
        "contribution": 0.02899562032303954
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.628909111022949,
        "normalized": 0.9644381976771178,
        "contribution": 0.14466572965156765
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.259568929672241,
        "normalized": 0.7674481251957699,
        "contribution": 0.07674481251957699
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c253a47aec06860c123f232e3bcc4fa526b239eef66420f8fc3a258b6af91bda",
    "id": "92c6bf29cf3e121a90fa03a38a85",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.432034492492676,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1281847059726715,
    "fitness": 0.7952417235647058,
    "legacy_fitness": 0.7952417235647058,
    "target_cdf": 0.9104853512991299,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6074790954589844,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6699948323182268,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6074790954589844,
        "normalized": 0.9104853512991299,
        "contribution": 0.45524267564956494
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8905830383300781,
        "normalized": 0.18034202015435835,
        "contribution": 0.04508550503858959
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1833531856536865,
        "normalized": 0.7037290732409538,
        "contribution": 0.10555936098614306
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8437211513519287,
        "normalized": 0.6410729064392915,
        "contribution": 0.06410729064392916
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e9621cccd363cfb4c41f090ad7d21b14c0fb7c16c78fdb5ecf4d45c5dab65b9a",
    "id": "92d14731a34262b46798f18e1732",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": 0.7367651462554932,
    "voiceclap": 0.03227337822318077,
    "fitness": 0.7613308307105187,
    "legacy_fitness": 0.7613308307105187,
    "target_cdf": 0.8729869637040064,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3964571952819824,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.642998888737365,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3964571952819824,
        "normalized": 0.8729869637040064,
        "contribution": 0.4364934818520032
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.707392930984497,
        "normalized": 0.130294529860791,
        "contribution": 0.03257363246519775
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.352372169494629,
        "normalized": 0.7528841549820324,
        "contribution": 0.11293262324730485
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7513105869293213,
        "normalized": 0.6099915117285923,
        "contribution": 0.060999151172859235
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "feb289efe57209253d4241d63d2312eb889c05aba847fe8166a2dce1b0bc0b8b",
    "id": "9302c3ecd2d2e65b3dafa35f7ed8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.2751336097717285,
    "wer": 0.03333333333333333,
    "identity": 0.7294777631759644,
    "voiceclap": -0.00041982531547546387,
    "fitness": 0.6777901354662487,
    "legacy_fitness": 0.6777901354662487,
    "target_cdf": 0.7262673179100112,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.836437702178955,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6052111338460535,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.836437702178955,
        "normalized": 0.7262673179100112,
        "contribution": 0.3631336589550056
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2400012016296387,
        "normalized": 0.3044239560639862,
        "contribution": 0.07610598901599655
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.018782138824463,
        "normalized": 0.6519627370457438,
        "contribution": 0.09779441055686157
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9691355228424072,
        "normalized": 0.6817707531818977,
        "contribution": 0.06817707531818977
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "dcf81893fa70a3af6c770d57e2013836c2d32f9d8e12b7df5f0a0e241f747ba1",
    "id": "931ea4f4867ba64554d22f47c9c6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 5.998785495758057,
    "genuineness": 3.986435890197754,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09262508153915405,
    "fitness": 0.8159677326649623,
    "legacy_fitness": 0.8159677326649623,
    "target_cdf": 0.9997931996922089,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.880023002624512,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.822948504335188,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.880023002624512,
        "normalized": 0.9997931996922089,
        "contribution": 0.49989659984610446
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.253570795059204,
        "normalized": 0.3099141419638045,
        "contribution": 0.07747853549095113
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.224923133850098,
        "normalized": 0.9900472697166317,
        "contribution": 0.14850709045749474
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.566519260406494,
        "normalized": 0.9706627854063763,
        "contribution": 0.09706627854063764
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ed01e4ca85a6fd617f099d4c63d7ace394e93518db7e0d46ecf26dc3c4a82c23",
    "id": "9377cb34fef6728c783878c0c45b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 3.9425034523010254,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.044038206338882446,
    "fitness": 0.8162914582707197,
    "legacy_fitness": 0.8162914582707197,
    "target_cdf": 0.9838478622074346,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4353346824645996,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7564117735484551,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4353346824645996,
        "normalized": 0.9838478622074346,
        "contribution": 0.4919239311037173
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7888107299804688,
        "normalized": 0.1512118069914083,
        "contribution": 0.03780295174785207
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.39762020111084,
        "normalized": 0.9453738037624609,
        "contribution": 0.14180607056436914
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5984079837799072,
        "normalized": 0.8487882013251657,
        "contribution": 0.08487882013251657
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e0fa537a1fabfa1edb28fed19ba8d26f700050f1cf70c8f01fabf2be5dd16681",
    "id": "93ae334690e279af72408e34197e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.55477237701416,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04287998750805855,
    "fitness": 0.8184287834997146,
    "legacy_fitness": 0.8184287834997146,
    "target_cdf": 0.9763929662774343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.273009777069092,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7073029545353494,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.273009777069092,
        "normalized": 0.9763929662774343,
        "contribution": 0.48819648313871716
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.490278720855713,
        "normalized": 0.08456572900066728,
        "contribution": 0.02114143225016682
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.903127908706665,
        "normalized": 0.6605858000030286,
        "contribution": 0.06605858000030286
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0415ed44e9e1d590efd52b091ce700b807f980b7ee7097c95fdbf583b6aaa7f6",
    "id": "93d79413e9ec7383def845b6ee6f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04162626713514328,
    "fitness": 0.814677071345255,
    "legacy_fitness": 0.814677071345255,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7429316686570693,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.826127290725708,
        "normalized": 0.16150748713874752,
        "contribution": 0.04037687178468688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9884166717529297,
        "normalized": 0.8929163005060587,
        "contribution": 0.1339374450759088
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3519797325134277,
        "normalized": 0.7917476313694349,
        "contribution": 0.0791747631369435
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4194368b7d3342738e35da7b6dd0cc2b85c0a6121d0cfac5dcb8cfc3511c86a3",
    "id": "945b4d790a2c06428c83d8ff0782",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": 0.7499737739562988,
    "voiceclap": -0.0661453902721405,
    "fitness": 0.7652435915245496,
    "legacy_fitness": 0.7652435915245496,
    "target_cdf": 0.8729869637040064,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3964571952819824,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6520210748058046,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3964571952819824,
        "normalized": 0.8729869637040064,
        "contribution": 0.4364934818520032
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8295196294784546,
        "normalized": 0.16246558721860938,
        "contribution": 0.040616396804652344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.290102005004883,
        "normalized": 0.7352947867084237,
        "contribution": 0.11029421800626356
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8591229915618896,
        "normalized": 0.6461697814288547,
        "contribution": 0.06461697814288547
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c6e4a01f7e8ef30b89c9e994bed108974fa2a5340ca874880e9eed2c318cb59c",
    "id": "946fbdfd61aac51016edcd7fb851",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.050724729895591736,
    "fitness": 0.8247740295116195,
    "legacy_fitness": 0.8247740295116195,
    "target_cdf": 0.9990117801396228,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.425514221191406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8067031521213367,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.425514221191406,
        "normalized": 0.9990117801396228,
        "contribution": 0.4995058900698114
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.124659299850464,
        "normalized": 0.2596155287980314,
        "contribution": 0.06490388219950785
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.0825910568237305,
        "normalized": 0.9862070216770844,
        "contribution": 0.14793105325156267
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.22327995300293,
        "normalized": 0.9436232660045483,
        "contribution": 0.09436232660045484
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "df2b62b08a58e6c1ae48f6145a97cba7067772739e5a3717b53a1852edef2eb2",
    "id": "9491c9e78772accb3d6a695c966e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.024092197418213,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.02461325377225876,
    "fitness": 0.823584527731738,
    "legacy_fitness": 0.823584527731738,
    "target_cdf": 0.9964005449628859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.003470420837402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.798172858603129,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.003470420837402,
        "normalized": 0.9964005449628859,
        "contribution": 0.49820027248144294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0907351970672607,
        "normalized": 0.24711589066920464,
        "contribution": 0.06177897266730116
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.593326568603516,
        "normalized": 0.9619219031491297,
        "contribution": 0.14428828547236947
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.179274559020996,
        "normalized": 0.939053279820155,
        "contribution": 0.09390532798201551
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "339e7c8de336d8ddc0b01565cf5e5e33b0f2eb5f886addb1ca0cffd5eee79a8b",
    "id": "94b863be7feb29dd513256e67556",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.936227321624756,
    "wer": 0.0,
    "identity": 0.565264105796814,
    "voiceclap": 0.016766726970672607,
    "fitness": 0.7689756451439316,
    "legacy_fitness": 0.7689756451439316,
    "target_cdf": 0.8929163174612232,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5019683837890625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6640923225032345,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5019683837890625,
        "normalized": 0.8929163174612232,
        "contribution": 0.4464581587306116
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.570317506790161,
        "normalized": 0.8092130613886472,
        "contribution": 0.12138195920829709
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9163296222686768,
        "normalized": 0.6648664054810712,
        "contribution": 0.06648664054810713
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d1a48f34bd7e72630dfc4110374ab0a32ec15dc27e47d152e553bbe78be0a693",
    "id": "94c66d2a1427fa164fe27b1523a9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 3.823258876800537,
    "wer": 0.0,
    "identity": 0.7793957591056824,
    "voiceclap": 0.04749511182308197,
    "fitness": 0.8038540289303425,
    "legacy_fitness": 0.8038540289303425,
    "target_cdf": 0.9772498618564534,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2892422676086426,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7604372327248263,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2892422676086426,
        "normalized": 0.9772498618564534,
        "contribution": 0.4886249309282267
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.917722225189209,
        "normalized": 0.18866878339787096,
        "contribution": 0.04716719584946774
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7040200233459473,
        "normalized": 0.8697054632450133,
        "contribution": 0.08697054632450134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a28e8e19cf56a9fe2ee9de5bfe1138fe9e92560b21c6df9c93e185d5e9a5b9e2",
    "id": "94d0b6b031988a5584d398e8fe2e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 5.959072113037109,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03999481722712517,
    "fitness": 0.8031552706373385,
    "legacy_fitness": 0.8031552706373385,
    "target_cdf": 0.9599408456112248,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0295231342315674,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8444051544420255,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0295231342315674,
        "normalized": 0.9599408456112248,
        "contribution": 0.4799704228056124
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.7721853256225586,
        "normalized": 0.5402542179334283,
        "contribution": 0.13506355448335708
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.08246374130249,
        "normalized": 0.9279837132182651,
        "contribution": 0.09279837132182651
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9024a5072bf6a0f9d425533ba9d50f8500001509861ef3b28e4dffcbcd41a57d",
    "id": "956724ab8669904688bd7e7d24f0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.05284466594457626,
    "fitness": 0.8352989265926694,
    "legacy_fitness": 0.8352989265926694,
    "target_cdf": 0.9964005449628859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.003470420837402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7788702839867325,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.003470420837402,
        "normalized": 0.9964005449628859,
        "contribution": 0.49820027248144294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.063595771789551,
        "normalized": 0.23735276923124937,
        "contribution": 0.05933819230781234
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.54120135307312,
        "normalized": 0.8365725957484663,
        "contribution": 0.08365725957484664
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6544ccc5e19c37ce697ee67af6389dc189228e8141317f203883412af39a5c45",
    "id": "95f13f7da03b371d704af9046382",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.773050308227539,
    "wer": 0.0,
    "identity": 0.6764127612113953,
    "voiceclap": 0.004594199359416962,
    "fitness": 0.7961636398564333,
    "legacy_fitness": 0.7961636398564333,
    "target_cdf": 0.9557276754241297,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.980825901031494,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6991283680744861,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.980825901031494,
        "normalized": 0.9557276754241297,
        "contribution": 0.47786383771206487
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4970635175704956,
        "normalized": 0.08578327395609947,
        "contribution": 0.021445818489024868
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.872772216796875,
        "normalized": 0.8729869940654369,
        "contribution": 0.13094804910981553
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.991138219833374,
        "normalized": 0.688706627635808,
        "contribution": 0.06887066276358081
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f5b3fe52f1171696abee8e722ca3a3f8bae6d2f8eabad9152e6e8eef15c3ab7f",
    "id": "964de6292e6351080ce0b36aaace",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.557555198669434,
    "wer": 0.0,
    "identity": 0.6285220384597778,
    "voiceclap": -0.10497675836086273,
    "fitness": 0.8430260732836581,
    "legacy_fitness": 0.8430260732836581,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7183627122476586,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.0239996910095215,
        "normalized": 0.8985685934084969,
        "contribution": 0.13478528901127454
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.878925323486328,
        "normalized": 0.6526844830621856,
        "contribution": 0.06526844830621857
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a87d9d89b0f40712f1b82b4253c4d2f58ef9b335a6f5ee8577c35aa5cffde426",
    "id": "96a348f497e7bfc0291a491a404f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.507347106933594,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.05511212348937988,
    "fitness": 0.8271186841725386,
    "legacy_fitness": 0.8271186841725386,
    "target_cdf": 0.9631976153848724,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0701041221618652,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7616263018331306,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0701041221618652,
        "normalized": 0.9631976153848724,
        "contribution": 0.4815988076924362
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.265444278717041,
        "normalized": 0.3147531709187549,
        "contribution": 0.07868829272968872
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5614216327667236,
        "normalized": 0.8070803373590341,
        "contribution": 0.12106205060385511
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.395984649658203,
        "normalized": 0.8027715080715057,
        "contribution": 0.08027715080715057
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "403fff270a561be13b604113e700a6b6f302181dbd5bf4b80ff16ed0afb1b1db",
    "id": "96c0f0f6737f65daccc21d1d3f3e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": 0.6266212463378906,
    "voiceclap": -0.1455918699502945,
    "fitness": 0.7644405099468036,
    "legacy_fitness": 0.7644405099468036,
    "target_cdf": 0.8450960105094518,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2665977478027344,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6685856781996392,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2665977478027344,
        "normalized": 0.8450960105094518,
        "contribution": 0.4225480052547259
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.307849407196045,
        "normalized": 0.33229177143790456,
        "contribution": 0.08307294285947614
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.9409446716308594,
        "normalized": 0.6263742746490999,
        "contribution": 0.09395614119736498
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9955387115478516,
        "normalized": 0.690085888880722,
        "contribution": 0.0690085888880722
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4cdd4e38d1d763a7742683ac3ea9ec419832e3c7cd1dd4035002011cea31e249",
    "id": "96ee67db6af8b08041c7a5a1e39e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.162164688110352,
    "wer": 0.0,
    "identity": 0.6490115523338318,
    "voiceclap": -0.10878604650497437,
    "fitness": 0.823779837788688,
    "legacy_fitness": 0.823779837788688,
    "target_cdf": 0.9872707933218218,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.532729148864746,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7248467282804685,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.532729148864746,
        "normalized": 0.9872707933218218,
        "contribution": 0.4936353966609109
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6666839122772217,
        "normalized": 0.12062361006755057,
        "contribution": 0.030155902516887642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.730440616607666,
        "normalized": 0.845096013305189,
        "contribution": 0.12676440199577835
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1715588569641113,
        "normalized": 0.7429102710689157,
        "contribution": 0.07429102710689157
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3250ef79851982480606584d150c083496f03a9b894b3b2284f2131021d9b071",
    "id": "972b6e9e41cf572935284ed76f21",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.01910388097167015,
    "fitness": 0.8401892810873963,
    "legacy_fitness": 0.8401892810873963,
    "target_cdf": 0.999110974830941,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.457979202270508,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7479081668006639,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.457979202270508,
        "normalized": 0.999110974830941,
        "contribution": 0.4995554874154705
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5784813165664673,
        "normalized": 0.10143141696818753,
        "contribution": 0.02535785424204688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.602221965789795,
        "normalized": 0.9625641895690078,
        "contribution": 0.14438462843535116
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.329977035522461,
        "normalized": 0.7861019670779525,
        "contribution": 0.07861019670779526
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4bd5b4973d26b5d3ade98dda5354e44aed910d355bb47e651d7c2914ac56dd5f",
    "id": "984a145e52a5caa839eb5f24f524",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.18615520000457764,
    "fitness": 0.8021785938000026,
    "legacy_fitness": 0.8021785938000026,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6847541741281213,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8668360710144043,
        "normalized": 0.17324834271715545,
        "contribution": 0.04331208567928886
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4101943969726562,
        "normalized": 0.7686399083613955,
        "contribution": 0.11529598625420932
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.903127908706665,
        "normalized": 0.6605858000030286,
        "contribution": 0.06605858000030286
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "74c811bc68a873d5455a59b467057975a945d8938d398eea5837ff07b4070cae",
    "id": "9853bf221557f24abadc48ddfd2f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.260893821716309,
    "genuineness": 3.998987913131714,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.011528141796588898,
    "fitness": 0.8192866755347155,
    "legacy_fitness": 0.8192866755347155,
    "target_cdf": 0.9922796672226393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7275185585021973,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7774673796587446,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7275185585021973,
        "normalized": 0.9922796672226393,
        "contribution": 0.49613983361131964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0194945335388184,
        "normalized": 0.22194973203149376,
        "contribution": 0.05548743300787344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.228601455688477,
        "normalized": 0.9269054845040974,
        "contribution": 0.1390358226756146
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.695219039916992,
        "normalized": 0.8680429036393686,
        "contribution": 0.08680429036393687
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bd9261da3f6259b75cc770dce921b707cf75329831c558d734ab5f09c93ccec2",
    "id": "98f68bd57a0fe2924de1b2ee0302",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 3.998987913131714,
    "wer": 0.0,
    "identity": 0.6646432280540466,
    "voiceclap": 0.005630733445286751,
    "fitness": 0.7897425013190451,
    "legacy_fitness": 0.7897425013190451,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6527975240836643,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5377724170684814,
        "normalized": 0.09336535255322709,
        "contribution": 0.023341338138306772
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.383507251739502,
        "normalized": 0.7614385460618895,
        "contribution": 0.11421578190928343
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.549987316131592,
        "normalized": 0.540060393868238,
        "contribution": 0.0540060393868238
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2c2a4da1cdbce6cf4f73c35e98a7449f4208190a05ca792098076cfca38539c3",
    "id": "98fc9530565545fe042134a5ed47",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.036643981933594,
    "wer": 0.0,
    "identity": 0.8227210640907288,
    "voiceclap": -0.07306787371635437,
    "fitness": 0.7362150569892485,
    "legacy_fitness": 0.7362150569892485,
    "target_cdf": 0.8186299522989113,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.1570286750793457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6171834587131957,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.1570286750793457,
        "normalized": 0.8186299522989113,
        "contribution": 0.40931497614945567
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9041526317596436,
        "normalized": 0.18447617999078308,
        "contribution": 0.04611904499769577
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.094395875930786,
        "normalized": 0.6761756814723885,
        "contribution": 0.10142635222085827
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.731508255004883,
        "normalized": 0.6032308534518596,
        "contribution": 0.06032308534518596
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "153f83416640f0dd38e3ba4647a578aa36b74e0134e5efdf8ba38ee1ff2da6c5",
    "id": "9926a6e16b20df2fc940a61c4354",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.287685871124268,
    "wer": 0.0,
    "identity": 0.5711437463760376,
    "voiceclap": 0.025458747521042824,
    "fitness": 0.7806289384043926,
    "legacy_fitness": 0.7806289384043926,
    "target_cdf": 0.8778006373398928,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4208059310913086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6814045768217354,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4208059310913086,
        "normalized": 0.8778006373398928,
        "contribution": 0.4389003186699464
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.10091233253479,
        "normalized": 0.2508316293237907,
        "contribution": 0.06270790733094768
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.285654067993164,
        "normalized": 0.7340144537731833,
        "contribution": 0.11010216806597749
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0175411701202393,
        "normalized": 0.6969418275486379,
        "contribution": 0.06969418275486379
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a7c4c8489facf8943d91ce81ab95ac7a83b711f292d04d4c51d55915a9233a54",
    "id": "99767851323788dbff039aa6827b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": 0.5344352722167969,
    "voiceclap": -0.04302040487527847,
    "fitness": 0.8245506224746298,
    "legacy_fitness": 0.8245506224746298,
    "target_cdf": 0.9977706984112247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.165794849395752,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7692250070361902,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.165794849395752,
        "normalized": 0.9977706984112247,
        "contribution": 0.49888534920561234
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.353141784667969,
        "normalized": 0.9409148986309699,
        "contribution": 0.1411372347946455
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.686418056488037,
        "normalized": 0.8663657697811189,
        "contribution": 0.08663657697811189
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cd7d87d4fbaca98494d7094fbe7582ed00f5a8180e1783d8916d87f943c7a156",
    "id": "998ba1cbfc334ee94479a723843a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": 0.7993521094322205,
    "voiceclap": -0.05003104731440544,
    "fitness": 0.7588002450759195,
    "legacy_fitness": 0.7588002450759195,
    "target_cdf": 0.8680429235090776,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3721086978912354,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6525937862352735,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3721086978912354,
        "normalized": 0.8680429235090776,
        "contribution": 0.4340214617545388
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3924028873443604,
        "normalized": 0.7638525190595127,
        "contribution": 0.1145778778589269
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.878925323486328,
        "normalized": 0.6526844830621856,
        "contribution": 0.06526844830621857
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6b7a6de912720d29d80cb9a6775386795ee2efa568f919e4c5a1925b94302365",
    "id": "99db1beed56f58c5f4ae42ba083c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": 0.5006859302520752,
    "voiceclap": -0.06135442107915878,
    "fitness": 0.8381860389739287,
    "legacy_fitness": 0.8381860389739287,
    "target_cdf": 0.9962285626788958,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9872379302978516,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7183836036974713,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9872379302978516,
        "normalized": 0.9962285626788958,
        "contribution": 0.4981142813394479
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4970635175704956,
        "normalized": 0.08578327395609947,
        "contribution": 0.021445818489024868
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.755711078643799,
        "normalized": 0.6114894424636795,
        "contribution": 0.06114894424636796
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f94f3fbf8eecb3f5a61e4ea9ea378a695aa5ef24f6b5f4764d11981baeae8de4",
    "id": "9b10ace9b3df4754911c73627eb3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.413206577301025,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04048587381839752,
    "fitness": 0.8310821984429613,
    "legacy_fitness": 0.8310821984429613,
    "target_cdf": 0.98254389250665,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.402869701385498,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7415564540605177,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.402869701385498,
        "normalized": 0.98254389250665,
        "contribution": 0.491271946253325
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7752411365509033,
        "normalized": 0.1475786358230735,
        "contribution": 0.03689465895576838
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8193979263305664,
        "normalized": 0.8629676950479268,
        "contribution": 0.12944515425718903
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5544028282165527,
        "normalized": 0.8394469459423535,
        "contribution": 0.08394469459423536
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8d661cf3d9272c7e2190d4a33a17eb8e194db702f46835f27e0116021641009f",
    "id": "9b21c794e1f50816292461934c94",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 4.281409740447998,
    "wer": 0.03333333333333333,
    "identity": 0.518794059753418,
    "voiceclap": -0.11904013156890869,
    "fitness": 0.7570990233916844,
    "legacy_fitness": 0.7570990233916844,
    "target_cdf": 0.8985686055474182,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.534433364868164,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7435921166306998,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.534433364868164,
        "normalized": 0.8985686055474182,
        "contribution": 0.4492843027737091
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.406229257583618,
        "normalized": 0.37436575159516905,
        "contribution": 0.09359143789879226
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.414642333984375,
        "normalized": 0.7698282582796319,
        "contribution": 0.11547423874194478
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6160099506378174,
        "normalized": 0.852421372162536,
        "contribution": 0.08524213721625361
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bcf43c8867b67fb5868f516afcddb8d4052476b64d8e8dd8e7165e0bcf0955da",
    "id": "9bb74c2de78db7fc7c447b4842fb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": 0.3879629969596863,
    "voiceclap": -0.0023274384438991547,
    "fitness": 0.8374043128878421,
    "legacy_fitness": 0.8374043128878421,
    "target_cdf": 0.9983456496378873,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.263189792633057,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7498775400902157,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.263189792633057,
        "normalized": 0.9983456496378873,
        "contribution": 0.49917282481894365
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.632759928703308,
        "normalized": 0.11295969157872005,
        "contribution": 0.02823992289468001
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.379828929901123,
        "normalized": 0.9436232726705875,
        "contribution": 0.14154349090058813
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4223875999450684,
        "normalized": 0.8092130147600393,
        "contribution": 0.08092130147600393
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b6415d42d798237f5b72ebc095615ffbdd99d022bfcb311089aec12485339c75",
    "id": "9c0011c2d322a83e1efbd96b7638",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.538887023925781,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.02530626580119133,
    "fitness": 0.8211696653401026,
    "legacy_fitness": 0.8211696653401026,
    "target_cdf": 0.9763929662774343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.273009777069092,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7559907883532604,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.273009777069092,
        "normalized": 0.9763929662774343,
        "contribution": 0.48819648313871716
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9787856340408325,
        "normalized": 0.2082523969786677,
        "contribution": 0.05206309924466693
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.952833652496338,
        "normalized": 0.8870402984992682,
        "contribution": 0.13305604477489022
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4971961975097656,
        "normalized": 0.8267516119498604,
        "contribution": 0.08267516119498604
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7a8b0d8c5c5cceb53a6378d5a1ea4e415780100b31344d3c7cdfc45d02810492",
    "id": "9dfed3fae205954e1a51b4838276",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": 0.5316587686538696,
    "voiceclap": -0.09950077533721924,
    "fitness": 0.7757655637188723,
    "legacy_fitness": 0.7757655637188723,
    "target_cdf": 0.905324255769247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.575014352798462,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6901695648678786,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.575014352798462,
        "normalized": 0.905324255769247,
        "contribution": 0.4526621278846235
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8973677158355713,
        "normalized": 0.18240174630185035,
        "contribution": 0.04560043657546259
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.427985668182373,
        "normalized": 0.7733726160960801,
        "contribution": 0.116005892414412
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2287654876708984,
        "normalized": 0.7590110799338052,
        "contribution": 0.07590110799338053
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b5fda421d693b2c026f7da763216743e226c19f6c182c943f28487bfa29e1593",
    "id": "9e213be1f74824b5517bd28b859e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": 0.7630114555358887,
    "voiceclap": -0.050612155348062515,
    "fitness": 0.7617794100597561,
    "legacy_fitness": 0.7617794100597561,
    "target_cdf": 0.8559956519117078,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3152952194213867,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6378251853598234,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3152952194213867,
        "normalized": 0.8559956519117078,
        "contribution": 0.4279978259558539
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3390283584594727,
        "normalized": 0.7491683218823387,
        "contribution": 0.1123752482823508
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6985044479370117,
        "normalized": 0.5918948114376616,
        "contribution": 0.05918948114376617
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "819f10e40761e83f300953660244ff29fe922ffa1e822645478a1a410c993a39",
    "id": "9e99a0d941dcd506d75f0e77b8ca",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.557555198669434,
    "wer": 0.0,
    "identity": 0.5854911804199219,
    "voiceclap": -0.12047551572322845,
    "fitness": 0.8415447252023549,
    "legacy_fitness": 0.8415447252023549,
    "target_cdf": 0.988262823627061,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5651941299438477,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7333984182640795,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5651941299438477,
        "normalized": 0.988262823627061,
        "contribution": 0.4941314118135305
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8329119682312012,
        "normalized": 0.1634273784096072,
        "contribution": 0.0408568446024018
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9779367446899414,
        "normalized": 0.6845529821840911,
        "contribution": 0.06845529821840911
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ac2b90882eb349f2b4eb69964169d81756bc46e5e64ce45f4873aa2d67de14a3",
    "id": "9ec89d6db173727ad6a912b6bb62",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": 0.657831072807312,
    "voiceclap": -0.13380008935928345,
    "fitness": 0.8311997628335385,
    "legacy_fitness": 0.8311997628335385,
    "target_cdf": 0.9937903356798943,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.808681011199951,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7629572768569417,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.808681011199951,
        "normalized": 0.9937903356798943,
        "contribution": 0.49689516783994714
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.046633720397949,
        "normalized": 0.23136007854264456,
        "contribution": 0.05784001963566114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3167755603790283,
        "normalized": 0.782672257515953,
        "contribution": 0.07826722575159531
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6ae9639332e6a8278ac79798af74a25a36228d70c5a506ab3640602321b72f9c",
    "id": "9ee2f1b82ae4d3fa2690e3bf9dc2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.041565582156181335,
    "fitness": 0.8264536007729316,
    "legacy_fitness": 0.8264536007729316,
    "target_cdf": 0.9867481280341037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5164966583251953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7248291852538585,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5164966583251953,
        "normalized": 0.9867481280341037,
        "contribution": 0.49337406401705186
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.58526611328125,
        "normalized": 0.10282367927847097,
        "contribution": 0.025705919819617742
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.1663312911987305,
        "normalized": 0.9190091380825917,
        "contribution": 0.13785137071238873
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.960334539413452,
        "normalized": 0.678978307048002,
        "contribution": 0.0678978307048002
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8e523f483f0e03397636c91540aaff6d8c95ae495339df9d37e175d6bd32159f",
    "id": "9ee8946fd1bbcbf9a7215c3c8713",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.02291685901582241,
    "fitness": 0.7883040381000047,
    "legacy_fitness": 0.7883040381000047,
    "target_cdf": 0.9079315965425183,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5912468433380127,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6463602354008309,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5912468433380127,
        "normalized": 0.9079315965425183,
        "contribution": 0.45396579827125916
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8498740196228027,
        "normalized": 0.16829178263767364,
        "contribution": 0.04207294565941841
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.125530958175659,
        "normalized": 0.6859401050066158,
        "contribution": 0.10289101575099237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.3640661239624023,
        "normalized": 0.4743047571916094,
        "contribution": 0.047430475719160946
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "072424e333611786518d2310916edb2c7ef1641265c55036888c683d90e1f529",
    "id": "9efd4e09f02dd9b6f722a1c484c6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.483288764953613,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": 0.6099071502685547,
    "voiceclap": -0.093616783618927,
    "fitness": 0.7327577672513188,
    "legacy_fitness": 0.7327577672513188,
    "target_cdf": 0.7939810280227386,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.063692092895508,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6088562335796928,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.063692092895508,
        "normalized": 0.7939810280227386,
        "contribution": 0.3969905140113693
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.975393295288086,
        "normalized": 0.20713393848934814,
        "contribution": 0.051783484622337034
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.9965429306030273,
        "normalized": 0.6447161427405166,
        "contribution": 0.09670742141107748
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.821718692779541,
        "normalized": 0.6337481353490892,
        "contribution": 0.06337481353490892
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ee6c0ccd2cb6a12282208dea581673ddc8fe7a969420590f5b1c66ee1d788d5a",
    "id": "9f0d77f1050534a39836c2db2380",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": 0.6388943791389465,
    "voiceclap": 0.15040016174316406,
    "fitness": 0.8090033600338314,
    "legacy_fitness": 0.8090033600338314,
    "target_cdf": 0.9964005449628859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.003470420837402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7798069536087641,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.003470420837402,
        "normalized": 0.9964005449628859,
        "contribution": 0.49820027248144294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9109374284744263,
        "normalized": 0.18656518874247485,
        "contribution": 0.04664129718561871
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.67338752746582,
        "normalized": 0.9673901905993718,
        "contribution": 0.14510852858990578
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.871239185333252,
        "normalized": 0.8985685535179664,
        "contribution": 0.08985685535179665
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "66b2ba624775d8f001262865eaa9fbfc60ce3d7d9a515c3430aa662df7f7ee38",
    "id": "9f2c0d072d20adfcb45edd3a9862",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06829310953617096,
    "fitness": 0.8239326264896063,
    "legacy_fitness": 0.8239326264896063,
    "target_cdf": 0.9726892523841197,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2080800533294678,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7274932680051429,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2080800533294678,
        "normalized": 0.9726892523841197,
        "contribution": 0.48634462619205987
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.250767946243286,
        "normalized": 0.765054436633194,
        "contribution": 0.0765054436633194
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0df594be8b2225dade42bec24c19e31fc952f60d4f41bb881cfacf9e87b97d8d",
    "id": "9f4039ace6bd49f703436f9cffd5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.080576419830322,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.004440762102603912,
    "fitness": 0.7917520659571828,
    "legacy_fitness": 0.7917520659571828,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6837761840439023,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4858078956604004,
        "normalized": 0.7883708195307028,
        "contribution": 0.11825562292960541
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.757911205291748,
        "normalized": 0.6122377475420941,
        "contribution": 0.06122377475420942
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "10cadce9e660a3052dcbaeb62be6dfd1087284042297e894000eb7ba50e8c6a9",
    "id": "9f4265641b28ea46496b1016c3ae",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.024776969105005264,
    "fitness": 0.8071625951614841,
    "legacy_fitness": 0.8071625951614841,
    "target_cdf": 0.9470810897522896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.891547203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7015183430610861,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.891547203063965,
        "normalized": 0.9470810897522896,
        "contribution": 0.4735405448761448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6281397342681885,
        "normalized": 0.8227201748759714,
        "contribution": 0.1234080262313957
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8965272903442383,
        "normalized": 0.6584377363795636,
        "contribution": 0.06584377363795636
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "89bc77863d61fbe3b80d63bf15bb9372223170b894efb010a83ddc4e42554ec8",
    "id": "9f90595eb95e4a602a5e63f15383",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.080576419830322,
    "wer": 0.0,
    "identity": 0.5770884156227112,
    "voiceclap": -0.020135480910539627,
    "fitness": 0.7557355701359023,
    "legacy_fitness": 0.7557355701359023,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6427045000702412,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2945499420166016,
        "normalized": 0.736571978107565,
        "contribution": 0.11048579671613475
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.911929130554199,
        "normalized": 0.663441906270323,
        "contribution": 0.0663441906270323
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1be8802991fd1ebb6834004d48e5154985431ef11e002a4967160f25803a21ea",
    "id": "9fae4f9a7fa5abb5fbc264ccdcf1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 3.5941832065582275,
    "wer": 0.0,
    "identity": 0.4574403762817383,
    "voiceclap": 0.12948626279830933,
    "fitness": 0.7978799086671128,
    "legacy_fitness": 0.7978799086671128,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7979785436505317,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2026846408843994,
        "normalized": 0.28955421544612325,
        "contribution": 0.07238855386153081
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.477681636810303,
        "normalized": 0.9527255009885056,
        "contribution": 0.14290882514827583
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.056060791015625,
        "normalized": 0.9247120351834592,
        "contribution": 0.09247120351834592
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "682e918e6ae92499443b3b2cda8d071510a4420aae3929eb1578508985ce5dca",
    "id": "a042c77f41a7a27def8cb1b90282",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.017816066741943,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.051469407975673676,
    "fitness": 0.8206940056742285,
    "legacy_fitness": 0.8206940056742285,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.69944165813711,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.3885064125061035,
        "normalized": 0.06782500036811512,
        "contribution": 0.01695625009202878
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.032895088195801,
        "normalized": 0.8999470264638179,
        "contribution": 0.13499205396957267
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.643498182296753,
        "normalized": 0.5728339295312939,
        "contribution": 0.05728339295312939
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c82401047589712340d834ba1d908729d1267bff6bfe83032828971c0708c17d",
    "id": "a0676a3c7b84115235faf3c13100",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 3.7291181087493896,
    "wer": 0.0,
    "identity": 0.6606486439704895,
    "voiceclap": 0.007920042611658573,
    "fitness": 0.787069425216141,
    "legacy_fitness": 0.787069425216141,
    "target_cdf": 0.9487457816418461,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9077796936035156,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7525357849600923,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9077796936035156,
        "normalized": 0.9487457816418461,
        "contribution": 0.47437289082092304
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.043241500854492,
        "normalized": 0.23017181441662055,
        "contribution": 0.057542953604155136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.997312545776367,
        "normalized": 0.8943502454552662,
        "contribution": 0.13415253681828992
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.677617073059082,
        "normalized": 0.8646740371672423,
        "contribution": 0.08646740371672423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2a1633ff2580de4bf77aebc822930b084cd2825d06392f2314d55ad9dbab1baa",
    "id": "a0b3e37789b03052fb9080387bf3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": 0.6278079748153687,
    "voiceclap": 0.007293134927749634,
    "fitness": 0.7670730647170515,
    "legacy_fitness": 0.7670730647170515,
    "target_cdf": 0.905324255769247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.575014352798462,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6633476412560722,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.575014352798462,
        "normalized": 0.905324255769247,
        "contribution": 0.4526621278846235
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4546730518341064,
        "normalized": 0.7803682683335784,
        "contribution": 0.11705524025003676
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.812917709350586,
        "normalized": 0.6308043511605621,
        "contribution": 0.06308043511605621
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e9473e152f403ce1b848b3109bf324b6c45eb63a80986638ae8232517ffec6f8",
    "id": "a1299adc5bf4a5bb5d6deadb1455",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": 0.6965157389640808,
    "voiceclap": -0.06705315411090851,
    "fitness": 0.7737232728072099,
    "legacy_fitness": 0.7737232728072099,
    "target_cdf": 0.9178304042951448,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6561765670776367,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6704503833046032,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6561765670776367,
        "normalized": 0.9178304042951448,
        "contribution": 0.4589152021475724
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.427985668182373,
        "normalized": 0.7733726160960801,
        "contribution": 0.116005892414412
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.870124340057373,
        "normalized": 0.6497945073726306,
        "contribution": 0.06497945073726306
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2e1968b492a93fb8961a916373e5f3c8cc87bad3feef64f6c0e104d19a95cc97",
    "id": "a1715394d3b09c99d29d46a65844",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.256305694580078,
    "wer": 0.06666666666666667,
    "identity": 0.407232403755188,
    "voiceclap": -0.07646825164556503,
    "fitness": 0.7659924951991154,
    "legacy_fitness": 0.7659924951991154,
    "target_cdf": 0.9772498618564534,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2892422676086426,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7744879430090224,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2892422676086426,
        "normalized": 0.9772498618564534,
        "contribution": 0.4886249309282267
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0839502811431885,
        "normalized": 0.24465512789027288,
        "contribution": 0.06116378197256822
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.07737398147583,
        "normalized": 0.9066346475271524,
        "contribution": 0.13599519712907285
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8008313179016113,
        "normalized": 0.8870403297915466,
        "contribution": 0.08870403297915466
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ae480aa74fe7a6cda6f654cfc9a41f71544760f0b00af581813da018c57ae731",
    "id": "a17a06f45c1ef1b533105a386b8f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": 0.4795669615268707,
    "voiceclap": -0.06518363952636719,
    "fitness": 0.8290034852287227,
    "legacy_fitness": 0.8290034852287227,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7218969039156987,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6414833068847656,
        "normalized": 0.8257492948479319,
        "contribution": 0.12386239422718978
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.863523483276367,
        "normalized": 0.6476212685529261,
        "contribution": 0.06476212685529262
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c06af03918b5054bd5bfaca895ab612d08e780a7fbedc31f5acea7c15c183931",
    "id": "a1aa2fe1b3ae5528a1664e37027f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.557555198669434,
    "wer": 0.0,
    "identity": 0.596331000328064,
    "voiceclap": -0.09940024465322495,
    "fitness": 0.8357436087169317,
    "legacy_fitness": 0.8357436087169317,
    "target_cdf": 0.9726892523841197,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2080800533294678,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6971860736307032,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2080800533294678,
        "normalized": 0.9726892523841197,
        "contribution": 0.48634462619205987
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5347344875335693,
        "normalized": 0.8005954285570323,
        "contribution": 0.12008931428355483
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7623119354248047,
        "normalized": 0.6137332671590532,
        "contribution": 0.06137332671590532
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7026cc8e7513482577d6432f9413bb69b9561b8987ec7de0aade4df04f26adbe",
    "id": "a220bc784fa1d56d58972dfae23a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": 0.5873994827270508,
    "voiceclap": 0.043427903205156326,
    "fitness": 0.8367374317251725,
    "legacy_fitness": 0.8367374317251725,
    "target_cdf": 0.9994229746665722,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.587838649749756,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7389671332155848,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.587838649749756,
        "normalized": 0.9994229746665722,
        "contribution": 0.4997114873332861
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.3545823097229004,
        "normalized": 0.0628538085933158,
        "contribution": 0.01571345214832895
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.611117362976074,
        "normalized": 0.9631976001981115,
        "contribution": 0.14447964002971672
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.347579002380371,
        "normalized": 0.790625537042531,
        "contribution": 0.07906255370425311
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "362e696499c407935b2969f58ef90afaf2e80a98c0ac20a1ff9ca3886b8e6f13",
    "id": "a2835326d69e374d54f4a830b0d1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12905235588550568,
    "fitness": 0.8253618587660804,
    "legacy_fitness": 0.8253618587660804,
    "target_cdf": 0.9964005449628859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.003470420837402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.810278543817343,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.003470420837402,
        "normalized": 0.9964005449628859,
        "contribution": 0.49820027248144294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2366087436676025,
        "normalized": 0.30305817864738066,
        "contribution": 0.07576454466184517
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.71786642074585,
        "normalized": 0.9701371067780673,
        "contribution": 0.1455205660167101
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9328465461730957,
        "normalized": 0.9079316065734483,
        "contribution": 0.09079316065734483
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "edbed7cffb265bcf9501aa228019005a41cc724d98991a01b876e5399fac4ceb",
    "id": "a2ccd9a9ef5c31c299909b69feed",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": 0.7814416885375977,
    "voiceclap": -0.05359862372279167,
    "fitness": 0.7770194260822667,
    "legacy_fitness": 0.7770194260822667,
    "target_cdf": 0.8855360207054879,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4613871574401855,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6560331344547152,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4613871574401855,
        "normalized": 0.8855360207054879,
        "contribution": 0.44276801035274393
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7820258140563965,
        "normalized": 0.14938781636553056,
        "contribution": 0.03734695409138264
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2767584323883057,
        "normalized": 0.7314444798958826,
        "contribution": 0.10971667198438238
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9075284004211426,
        "normalized": 0.6620149802620625,
        "contribution": 0.06620149802620626
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cb369a2737f0581cdbb80f713b07e3c3002da2096df8ced16b32f304ba6ea5ae",
    "id": "a2da338a6719f2dd09454294e6ff",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": 0.6316522359848022,
    "voiceclap": -0.11154929548501968,
    "fitness": 0.7294668588069517,
    "legacy_fitness": 0.7294668588069517,
    "target_cdf": 0.8016852523937701,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.0920987129211426,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5996069689671766,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.0920987129211426,
        "normalized": 0.8016852523937701,
        "contribution": 0.40084262619688504
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8973677158355713,
        "normalized": 0.18240174630185035,
        "contribution": 0.04560043657546259
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.974303722381592,
        "normalized": 0.637416836817112,
        "contribution": 0.0956125255225668
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6511991024017334,
        "normalized": 0.5755138067226213,
        "contribution": 0.057551380672262134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6308d57cce58b86fbe72eb584f94a10bf14715b5a37f03ddfca0ed3b3328bb20",
    "id": "a2f8db1347000b42a5b93b106840",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 6.117925643920898,
    "genuineness": 3.7793264389038086,
    "wer": 0.0,
    "identity": 0.6903795003890991,
    "voiceclap": 0.11489808559417725,
    "fitness": 0.8103166758984522,
    "legacy_fitness": 0.8103166758984522,
    "target_cdf": 0.9997931996922089,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.880023002624512,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8137445494417528,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.880023002624512,
        "normalized": 0.9997931996922089,
        "contribution": 0.49989659984610446
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1517984867095947,
        "normalized": 0.26984519049493305,
        "contribution": 0.06746129762373326
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.31387996673584,
        "normalized": 0.9919413784114786,
        "contribution": 0.14879120676172178
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.663330554962158,
        "normalized": 0.975954452101933,
        "contribution": 0.09759544521019331
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fcd6193732272040c7303e17a2d2757c6ac78e23e12a450bb93363bc0f573020",
    "id": "a352bc5100e6702cf5a47dc41a17",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.607763767242432,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1001501977443695,
    "fitness": 0.7968790250190385,
    "legacy_fitness": 0.7968790250190385,
    "target_cdf": 0.8943502420521354,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.510084629058838,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6959300691884074,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.510084629058838,
        "normalized": 0.8943502420521354,
        "contribution": 0.4471751210260677
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.131443977355957,
        "normalized": 0.2621540050393544,
        "contribution": 0.0655385012598386
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1878011226654053,
        "normalized": 0.7050781385006836,
        "contribution": 0.10576172077510253
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2859721183776855,
        "normalized": 0.7745472612739859,
        "contribution": 0.0774547261273986
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cb429f0b5621b45f6888c116f6c54bebd3b8070628fe46f2d9a5619ccb3ec155",
    "id": "a4b130e66c0c979c3090794199ae",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04138464108109474,
    "fitness": 0.7995678005445913,
    "legacy_fitness": 0.7995678005445913,
    "target_cdf": 0.9519500370480449,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.940244674682617,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7025230211777391,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.940244674682617,
        "normalized": 0.9519500370480449,
        "contribution": 0.47597501852402246
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7838149070739746,
        "normalized": 0.8559956150601711,
        "contribution": 0.12839934225902566
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.900927782058716,
        "normalized": 0.6598703799739153,
        "contribution": 0.06598703799739153
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a335e9016f29b2e8306399c601d0881dedfd05416c8267934aa366ea7533df0e",
    "id": "a54226fd36f9977d6025013cb803",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.19130755960941315,
    "fitness": 0.8307415980224655,
    "legacy_fitness": 0.8307415980224655,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7273041662222635,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.598835825920105,
        "normalized": 0.10564978306864803,
        "contribution": 0.026412445767162007
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.903127908706665,
        "normalized": 0.6605858000030286,
        "contribution": 0.06605858000030286
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0b94ba00f6a41d866035b754c9a69f0c0eeca6ff742c6dda4131d8f19ee55222",
    "id": "a55cd731f4f8cc7810db5930644e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.515059471130371,
    "genuineness": 4.080576419830322,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.16414162516593933,
    "fitness": 0.8305271156430589,
    "legacy_fitness": 0.8305271156430589,
    "target_cdf": 0.9952532227437393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9060754776000977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7165369699392349,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9060754776000977,
        "normalized": 0.9952532227437393,
        "contribution": 0.49762661137186964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4970635175704956,
        "normalized": 0.08578327395609947,
        "contribution": 0.021445818489024868
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.184122562408447,
        "normalized": 0.9213282021786904,
        "contribution": 0.13819923032680356
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.70070481300354,
        "normalized": 0.5926530975153677,
        "contribution": 0.059265309751536766
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "50bb15ddc78b8c63f78fda4a0d10556ba2f8fe5dd2cd462502fa1ca4bbdfec9c",
    "id": "a5742183b882edd8667f671e3640",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.8420867919921875,
    "wer": 0.0,
    "identity": 0.482004851102829,
    "voiceclap": 0.04071595519781113,
    "fitness": 0.801892935011836,
    "legacy_fitness": 0.801892935011836,
    "target_cdf": 0.9673902089874984,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.126917839050293,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7134183822672066,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.126917839050293,
        "normalized": 0.9673902089874984,
        "contribution": 0.4836951044937492
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6056206226348877,
        "normalized": 0.10708370011583324,
        "contribution": 0.02677092502895831
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.854980707168579,
        "normalized": 0.8697054717721656,
        "contribution": 0.13045582076582485
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1099517345428467,
        "normalized": 0.7249653197867423,
        "contribution": 0.07249653197867423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6555aa449df9b8702a9037dd0143b9b0d4bdc7f524d64ac97bf2ef24c30758eb",
    "id": "a6fa2eda97fbc031ee3fa0eb2ab0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.618313789367676,
    "genuineness": 4.545003414154053,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.27374526858329773,
    "fitness": 0.8137481891381831,
    "legacy_fitness": 0.8137481891381831,
    "target_cdf": 0.9178304042951448,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6561765670776367,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6572735976863552,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6561765670776367,
        "normalized": 0.9178304042951448,
        "contribution": 0.4589152021475724
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6191902160644531,
        "normalized": 0.1099935309435226,
        "contribution": 0.02749838273588065
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3568201065063477,
        "normalized": 0.7541161908392326,
        "contribution": 0.11311742862588489
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6566996574401855,
        "normalized": 0.5774258417701726,
        "contribution": 0.05774258417701727
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f945978e3eae9db2bd2915e464f086fce2235f0e9cf1b1b9d0263bc82ec32a4b",
    "id": "a70c4e5875ebf788c6bfaeb85403",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.02406461164355278,
    "fitness": 0.8375146283351691,
    "legacy_fitness": 0.8375146283351691,
    "target_cdf": 0.9992824581550183,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.522909164428711,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7616742892479854,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.522909164428711,
        "normalized": 0.9992824581550183,
        "contribution": 0.49964122907750913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.646700859069824,
        "normalized": 0.9656443012467251,
        "contribution": 0.14484664518700877
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.686418056488037,
        "normalized": 0.8663657697811189,
        "contribution": 0.08663657697811189
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c8f85366481a29673b90146b94737fa9476687c96e9b2e9bb495f2fc0324d5b0",
    "id": "a724d7a66d20f9fc22c96876f6a5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.469690799713135,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10883885622024536,
    "fitness": 0.8385315858994193,
    "legacy_fitness": 0.8385315858994193,
    "target_cdf": 0.9919413812308048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7112860679626465,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7360092502521923,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7112860679626465,
        "normalized": 0.9919413812308048,
        "contribution": 0.4959706906154024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.693823218345642,
        "normalized": 0.127012998991818,
        "contribution": 0.0317532497479545
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.872772216796875,
        "normalized": 0.8729869940654369,
        "contribution": 0.13094804910981553
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.281571388244629,
        "normalized": 0.7733726077901995,
        "contribution": 0.07733726077901995
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "43e3b4fe690a1c90aeb41ffed218c0bbd2962eaee52c7ca0a469ac22918aefc3",
    "id": "a7ad1842ca2df9da83649a8c2f5d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.400654315948486,
    "wer": 0.07142857142857142,
    "identity": null,
    "voiceclap": -0.09951341897249222,
    "fitness": 0.7077050039278799,
    "legacy_fitness": 0.7077050039278799,
    "target_cdf": 0.8413447460685429,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2503652572631836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6111909629037635,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2503652572631836,
        "normalized": 0.8413447460685429,
        "contribution": 0.42067237303427146
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.982177972793579,
        "normalized": 0.2093744108235418,
        "contribution": 0.05234360270588545
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.6040189266204834,
        "normalized": 0.5105177719159082,
        "contribution": 0.07657766578738623
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7689125537872314,
        "normalized": 0.6159732137622032,
        "contribution": 0.061597321376220315
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4b78407bd4ced824722b54463ffcb187ff3cd13b9b4cf6c8ee0adc74f4ce2579",
    "id": "a7aee7a1fd35197535a4e335354e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09377911686897278,
    "fitness": 0.8234671768979988,
    "legacy_fitness": 0.8234671768979988,
    "target_cdf": 0.9981710548076552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.230724811553955,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8759390812749387,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.230724811553955,
        "normalized": 0.9981710548076552,
        "contribution": 0.4990855274038276
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.7335968017578125,
        "normalized": 0.5225843140407042,
        "contribution": 0.13064607851017604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.2427144050598145,
        "normalized": 0.9904545174589181,
        "contribution": 0.1485681776188377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.672131538391113,
        "normalized": 0.9763929774209734,
        "contribution": 0.09763929774209734
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "515120cadbc2f51fd4cbecfb9e8c5442f98a1b82d2811dab507d83904beb3402",
    "id": "a7e3eca13d76704d80e57a314e56",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 4.532451152801514,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1346314549446106,
    "fitness": 0.7847668497311009,
    "legacy_fitness": 0.7847668497311009,
    "target_cdf": 0.8855360207054879,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4613871574401855,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7001498278212402,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4613871574401855,
        "normalized": 0.8855360207054879,
        "contribution": 0.44276801035274393
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.195899724960327,
        "normalized": 0.28688766529430837,
        "contribution": 0.07172191632357709
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.383507251739502,
        "normalized": 0.7614385460618895,
        "contribution": 0.11421578190928343
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0747478008270264,
        "normalized": 0.7144411923563575,
        "contribution": 0.07144411923563575
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6ed2734a487ef6ad18a97ec883c8daecb2cbe09475fe6b74aee965069282dd9f",
    "id": "a80b519e3fdddb1b4c32e3b2a91a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.948779582977295,
    "wer": 0.0,
    "identity": 0.5331624746322632,
    "voiceclap": -0.12611782550811768,
    "fitness": 0.7240754545570096,
    "legacy_fitness": 0.7240754545570096,
    "target_cdf": 0.7973043085814874,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.075866222381592,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5910816983942047,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.075866222381592,
        "normalized": 0.7973043085814874,
        "contribution": 0.3986521542907437
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.112187385559082,
        "normalized": 0.6817707678889575,
        "contribution": 0.10226561518334362
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.733708620071411,
        "normalized": 0.6039836490389876,
        "contribution": 0.060398364903898764
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e2f507c58a803e685773bae40f80641ff15bbdb8483f34a25ef5d35f46cc35b3",
    "id": "a83c526ad50c25f576cfa6ae1906",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12491412460803986,
    "fitness": 0.8221055005322451,
    "legacy_fitness": 0.8221055005322451,
    "target_cdf": 0.9557276754241297,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.980825901031494,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7455573243085287,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.980825901031494,
        "normalized": 0.9557276754241297,
        "contribution": 0.47786383771206487
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0805578231811523,
        "normalized": 0.24342971276978492,
        "contribution": 0.06085742819244623
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6414833068847656,
        "normalized": 0.8257492948479319,
        "contribution": 0.12386239422718978
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5103979110717773,
        "normalized": 0.8297366417682784,
        "contribution": 0.08297366417682785
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3ab6385bca8f767ee91b1f1ad591e8891f351d84e1caced71da3aaad9c49c13e",
    "id": "a88bb6fe3c2698cce68bc01c61db",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": 0.5740950107574463,
    "voiceclap": -0.05372098833322525,
    "fitness": 0.7598371849073179,
    "legacy_fitness": 0.7598371849073179,
    "target_cdf": 0.8375344088898514,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.234132766723633,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6030360406377268,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.234132766723633,
        "normalized": 0.8375344088898514,
        "contribution": 0.4187672044449257
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8608829975128174,
        "normalized": 0.599461323887337,
        "contribution": 0.08991919858310055
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6996045112609863,
        "normalized": 0.5922739566056935,
        "contribution": 0.05922739566056935
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "07fd8a835ea034103ff3c00ded8ff79cd8a3a6e5e245cc118cd10d6df0554eae",
    "id": "a9013de7806f52ca275bfe75c86d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06398652493953705,
    "fitness": 0.8318540188076808,
    "legacy_fitness": 0.8318540188076808,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7130110192250823,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7616714239120483,
        "normalized": 0.1440043780288135,
        "contribution": 0.036001094507203374
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.508047342300415,
        "normalized": 0.7939810079308474,
        "contribution": 0.1190971511896271
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9317312240600586,
        "normalized": 0.6698333561249803,
        "contribution": 0.06698333561249803
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f4b1595ce2a5c855acec5c55f549673a3f4060f20a0b3cde5eef47d8e7fdf281",
    "id": "a921ac7839da223b573cfc87992b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 3.773050308227539,
    "wer": 0.0,
    "identity": 0.7536119818687439,
    "voiceclap": 0.038832396268844604,
    "fitness": 0.8045653321071519,
    "legacy_fitness": 0.8045653321071519,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7451214713274402,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.353141784667969,
        "normalized": 0.9409148986309699,
        "contribution": 0.1411372347946455
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4487907886505127,
        "normalized": 0.81552382833801,
        "contribution": 0.08155238283380101
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a89f71099f419c25fdb8a51a033010d9c971c6e5aa33688786c6384446e1d42b",
    "id": "a9229f39f21567fd692e0b9b86c0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.53094482421875,
    "genuineness": 4.268857479095459,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07234859466552734,
    "fitness": 0.8092303603213313,
    "legacy_fitness": 0.8092303603213313,
    "target_cdf": 0.9361753561737703,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7941524982452393,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6690526416022275,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7941524982452393,
        "normalized": 0.9361753561737703,
        "contribution": 0.4680876780868852
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2144882678985596,
        "normalized": 0.7131122892698544,
        "contribution": 0.10696684339047816
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8239188194274902,
        "normalized": 0.634482821195086,
        "contribution": 0.0634482821195086
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0bf7dd27bac90e13f13bb44780614baa3aa3f41f0d75ce037c75353dc5e907cb",
    "id": "a9d72316777ab47ce500b81072a2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.287685871124268,
    "wer": 0.0,
    "identity": 0.24563246965408325,
    "voiceclap": -0.12192118167877197,
    "fitness": 0.8287116345791351,
    "legacy_fitness": 0.8287116345791351,
    "target_cdf": 0.988262823627061,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5651941299438477,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8021838325416742,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5651941299438477,
        "normalized": 0.988262823627061,
        "contribution": 0.4941314118135305
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2433934211730957,
        "normalized": 0.3057923700274824,
        "contribution": 0.0764480925068706
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.246392726898193,
        "normalized": 0.9290496824330181,
        "contribution": 0.1393574523649527
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.038458824157715,
        "normalized": 0.922468758563205,
        "contribution": 0.0922468758563205
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "26fff4f385a20fc5d2ccb2404ecd5db32faf51498264dbc7e91ab8eda8149423",
    "id": "a9ddb73d23f8fcb61b1ddbfea689",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 3.4341442584991455,
    "wer": 0.0,
    "identity": 0.7868798971176147,
    "voiceclap": 0.06087975203990936,
    "fitness": 0.7961129521984536,
    "legacy_fitness": 0.7961129521984536,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7934061024807656,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.03645658493042,
        "normalized": 0.2278053936557905,
        "contribution": 0.056951348413947625
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.886885643005371,
        "normalized": 0.97888519494114,
        "contribution": 0.146832779241171
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0e6310d6655c46fc525409035ed9c312ee77d957dec88d981fef35e9df1840a6",
    "id": "a9f09fe78b050a336788864abd12",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.18726921081543,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03752553462982178,
    "fitness": 0.8298099088470487,
    "legacy_fitness": 0.8298099088470487,
    "target_cdf": 0.9952532227437393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9060754776000977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8503027037633032,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9060754776000977,
        "normalized": 0.9952532227437393,
        "contribution": 0.49762661137186964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5775458812713623,
        "normalized": 0.4510349972390877,
        "contribution": 0.11275874930977192
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.566638946533203,
        "normalized": 0.9599408341035309,
        "contribution": 0.14399112511552964
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.3993000984191895,
        "normalized": 0.9592621796613195,
        "contribution": 0.09592621796613196
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "77c78d8d042acbadf508781009747b1d02272bb466e05484fcf0487b2ff164c7",
    "id": "aa22d24500f58a0c861f30e5e946",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": 0.8116912245750427,
    "voiceclap": 0.004571188241243362,
    "fitness": 0.8143194706977528,
    "legacy_fitness": 0.8143194706977528,
    "target_cdf": 0.9741313242397945,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.232429027557373,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7164818269184259,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.232429027557373,
        "normalized": 0.9741313242397945,
        "contribution": 0.48706566211989727
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5784813165664673,
        "normalized": 0.10143141696818753,
        "contribution": 0.02535785424204688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.890563726425171,
        "normalized": 0.8762105512752956,
        "contribution": 0.13143158269129432
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.114352226257324,
        "normalized": 0.7262672786518737,
        "contribution": 0.07262672786518737
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "253f5148b9404b42103d7aefe9f644a7bc7f20ba8c218ce518eedfac49fbd281",
    "id": "aa45206cf5324d16fd41c9469cee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.532451152801514,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12249159067869186,
    "fitness": 0.8107485533246996,
    "legacy_fitness": 0.8107485533246996,
    "target_cdf": 0.9311452866573339,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7535715103149414,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6949355861391685,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7535715103149414,
        "normalized": 0.9311452866573339,
        "contribution": 0.46557264332866694
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.063595771789551,
        "normalized": 0.23735276923124937,
        "contribution": 0.05933819230781234
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2456233501434326,
        "normalized": 0.7223522963184019,
        "contribution": 0.10835284444776029
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7711129188537598,
        "normalized": 0.6167190605492896,
        "contribution": 0.06167190605492896
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1fd3eda8d00dcf3b237fbea0adf5c9e2e9e26883414fa7f56b92c22866cce3d0",
    "id": "aa8fafc1d42e3f031e46b2820af6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.8,
    "cfg": 3,
    "generation": null,
    "ce": 6.0146708488464355,
    "genuineness": 3.785602569580078,
    "wer": 0.03333333333333333,
    "identity": 0.5142359137535095,
    "voiceclap": 0.08600474148988724,
    "fitness": 0.7795035078124138,
    "legacy_fitness": 0.7795035078124138,
    "target_cdf": 0.9965654549919187,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.019702911376953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8869970370351011,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.019702911376953,
        "normalized": 0.9965654549919187,
        "contribution": 0.49828272749595937
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.8404576778411865,
        "normalized": 0.571301088250857,
        "contribution": 0.14282527206271425
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.064799785614014,
        "normalized": 0.9856469755282963,
        "contribution": 0.14784704632924445
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.760141372680664,
        "normalized": 0.9804199114718302,
        "contribution": 0.09804199114718302
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6d5b6491608a3ec06a1149efa007b328a6f12a826e99d9e3fb5538a7d75dfd03",
    "id": "aac9606c4aae84ba972e3a4213e0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.018277347087860107,
    "fitness": 0.7861101212141889,
    "legacy_fitness": 0.7861101212141889,
    "target_cdf": 0.912986084230339,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.623711585998535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6657891412148809,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.623711585998535,
        "normalized": 0.912986084230339,
        "contribution": 0.4564930421151695
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.223383903503418,
        "normalized": 0.7157670812548709,
        "contribution": 0.10736506218823062
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.830519676208496,
        "normalized": 0.6366840693362819,
        "contribution": 0.06366840693362819
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "62f8e1e7d2ba01bd3d7b06e56f73d234d37dc61cea5f6edb68ef55b14d72d439",
    "id": "ab011d45654148e10dbe314d8598",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": 0.5635367631912231,
    "voiceclap": 0.0017552003264427185,
    "fitness": 0.8387797093516627,
    "legacy_fitness": 0.8387797093516627,
    "target_cdf": 0.9950331004711068,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.889843463897705,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.762994606196614,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.889843463897705,
        "normalized": 0.9950331004711068,
        "contribution": 0.4975165502355534
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.887190580368042,
        "normalized": 0.17931761586218503,
        "contribution": 0.04482940396554626
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.255288600921631,
        "normalized": 0.9301035297844544,
        "contribution": 0.13951552946766815
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4311888217926025,
        "normalized": 0.8113312252784615,
        "contribution": 0.08113312252784616
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7f06a35551e0f28cf99d6a9cceabd32c1f52ece928585b7bc5189f0f13e2a2fb",
    "id": "ab4a944b209c5543f8b8defd9199",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.337893962860107,
    "wer": 0.0,
    "identity": 0.684097170829773,
    "voiceclap": 0.06557595729827881,
    "fitness": 0.835275125599279,
    "legacy_fitness": 0.835275125599279,
    "target_cdf": 0.9904545229601343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6463565826416016,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.780256547917755,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6463565826416016,
        "normalized": 0.9904545229601343,
        "contribution": 0.4952272614800671
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.104304790496826,
        "normalized": 0.2520767788636856,
        "contribution": 0.0630191947159214
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.139644145965576,
        "normalized": 0.9154342857630313,
        "contribution": 0.1373151428644547
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.589607000350952,
        "normalized": 0.8469494885731179,
        "contribution": 0.0846949488573118
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "82181ee064fec899a5225d8c47b75070dd18caba909b6de8b6d5d16693cfcb0e",
    "id": "ab601b776ee38439e193ade741ca",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.149696350097656,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": 0.3326086103916168,
    "voiceclap": -0.15379470586776733,
    "fitness": 0.8050478388882677,
    "legacy_fitness": 0.8050478388882677,
    "target_cdf": 0.947918719056646,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8996634483337402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.709688524433682,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8996634483337402,
        "normalized": 0.947918719056646,
        "contribution": 0.473959359528323
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9652159214019775,
        "normalized": 0.20379983538139917,
        "contribution": 0.050949958845349794
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.30344557762146,
        "normalized": 0.7391168214554646,
        "contribution": 0.11086752321831969
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1583573818206787,
        "normalized": 0.7391168284168952,
        "contribution": 0.07391168284168952
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "da2c4f61eb6c52d84c17f03c475277e9325f9c8b4ba5c64bce6828774246fad8",
    "id": "abbec427dd46b213fed0539ef074",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": 0.5957126617431641,
    "voiceclap": 0.057429999113082886,
    "fitness": 0.8167496720084946,
    "legacy_fitness": 0.8167496720084946,
    "target_cdf": 0.9998163301009735,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.912487983703613,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7681503376735948,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.912487983703613,
        "normalized": 0.9998163301009735,
        "contribution": 0.49990816505048674
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6802536249160767,
        "normalized": 0.12378946157536314,
        "contribution": 0.030947365393840784
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.011425495147705,
        "normalized": 0.9838478542017792,
        "contribution": 0.14757717813026688
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.862438201904297,
        "normalized": 0.897176290990003,
        "contribution": 0.0897176290990003
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "63842b8cd820849d28062ff41950b3f6dbe433b9a9061147693d0ea3d0348d1a",
    "id": "abc3420a0e653f6d82a6569fcb0c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.038498878479004,
    "genuineness": 4.362998008728027,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13031208515167236,
    "fitness": 0.7799291766850439,
    "legacy_fitness": 0.7799291766850439,
    "target_cdf": 0.8943502420521354,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.510084629058838,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.714661443714768,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.510084629058838,
        "normalized": 0.8943502420521354,
        "contribution": 0.4471751210260677
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1280517578125,
        "normalized": 0.2608832234596053,
        "contribution": 0.06522080586490132
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6592745780944824,
        "normalized": 0.8297365948158377,
        "contribution": 0.12446048922237565
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.299173593521118,
        "normalized": 0.7780502760142334,
        "contribution": 0.07780502760142334
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c94d247e364b03a13419e4714c4294608a733c50ab6eae2f8b07637907a781a4",
    "id": "ac01091e7610c846e592e3ab91a8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.923675537109375,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09174288064241409,
    "fitness": 0.7924519250625126,
    "legacy_fitness": 0.7924519250625126,
    "target_cdf": 0.9409148593413338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.834733486175537,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7228074661930812,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.834733486175537,
        "normalized": 0.9409148593413338,
        "contribution": 0.4704574296706669
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.012709617614746,
        "normalized": 0.2196317185230387,
        "contribution": 0.054907929630759675
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5480780601501465,
        "normalized": 0.8038541354521456,
        "contribution": 0.12057812031782184
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2639694213867188,
        "normalized": 0.7686398657383282,
        "contribution": 0.07686398657383282
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9068563aba5863f4d4b00e831c7d08132d566a5403f70eafd2b9304f9d6f78ec",
    "id": "ac0ee58dbe62afb2ff22384fe05a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": 0.5953351259231567,
    "voiceclap": -0.12575596570968628,
    "fitness": 0.7608201999894688,
    "legacy_fitness": 0.7608201999894688,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6458263696713542,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9109374284744263,
        "normalized": 0.18656518874247485,
        "contribution": 0.04664129718561871
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2589669227600098,
        "normalized": 0.7262672960496188,
        "contribution": 0.10894009440744282
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8415210247039795,
        "normalized": 0.640342746822993,
        "contribution": 0.0640342746822993
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fced28e4003ffd4cfb7cd107213f35594bc6b93b65ef4743c22380f52d64e17b",
    "id": "ac15bb995b087c917ab4cd801935",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.012958556413650513,
    "fitness": 0.8174115465824376,
    "legacy_fitness": 0.8174115465824376,
    "target_cdf": 0.9696036431679269,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1593828201293945,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7143364036336153,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1593828201293945,
        "normalized": 0.9696036431679269,
        "contribution": 0.48480182158396345
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.828293561935425,
        "normalized": 0.8646740370257361,
        "contribution": 0.1297011055538604
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0043396949768066,
        "normalized": 0.692836384904358,
        "contribution": 0.0692836384904358
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "324e5778072a5e2955a00feb77500965e6bfe784c09fd171c113fe7aafd2011c",
    "id": "ac1dd652eed104d1885195c44365",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.356721878051758,
    "wer": 0.06666666666666667,
    "identity": 0.4317573010921478,
    "voiceclap": -0.1554652452468872,
    "fitness": 0.7676253735835791,
    "legacy_fitness": 0.7676253735835791,
    "target_cdf": 0.9644381951325223,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.086336612701416,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7307678654953778,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.086336612701416,
        "normalized": 0.9644381951325223,
        "contribution": 0.48221909756626113
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9211146831512451,
        "normalized": 0.1897260605487444,
        "contribution": 0.0474315151371861
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8460850715637207,
        "normalized": 0.8680429242459118,
        "contribution": 0.13020643863688677
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.057145595550537,
        "normalized": 0.7091081415504378,
        "contribution": 0.07091081415504379
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4c5edd96448ac730851b2af3a2975ced1e7ad66e412aca22d88f154c3bcfa0b8",
    "id": "ac7fb2b40e0748f98f9bca5eddff",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.331617832183838,
    "wer": 0.030303030303030304,
    "identity": 0.5559736490249634,
    "voiceclap": -0.06721442937850952,
    "fitness": 0.7825282160715711,
    "legacy_fitness": 0.7825282160715711,
    "target_cdf": 0.9427315386702796,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.850965976715088,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7429120809765468,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.850965976715088,
        "normalized": 0.9427315386702796,
        "contribution": 0.4713657693351398
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3010644912719727,
        "normalized": 0.32945945286060696,
        "contribution": 0.08236486321515174
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4190900325775146,
        "normalized": 0.7710131225645452,
        "contribution": 0.11565196838468178
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.145155906677246,
        "normalized": 0.7352948004157347,
        "contribution": 0.07352948004157348
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2f0604c91b24be2f77081144d53d97f54e10959e247fd013a28360bcea1cddb2",
    "id": "acb1ddd435ac0368d394cc268838",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 4.3881025314331055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.02267078310251236,
    "fitness": 0.8349742446053762,
    "legacy_fitness": 0.8349742446053762,
    "target_cdf": 0.9979798622347391,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1982598304748535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7257240438663395,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1982598304748535,
        "normalized": 0.9979798622347391,
        "contribution": 0.49898993111736956
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.3952912092208862,
        "normalized": 0.06885472951265481,
        "contribution": 0.017213682378163703
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.273079872131348,
        "normalized": 0.932174984106283,
        "contribution": 0.13982624761594245
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0175411701202393,
        "normalized": 0.6969418275486379,
        "contribution": 0.06969418275486379
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c04460d530402a5cd72d05b27961db3ddccdc34df9f6b5ffdd7501ff1109765a",
    "id": "acc53b59c0d12c37ae8fc84a61de",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 3.986435890197754,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.1633162796497345,
    "fitness": 0.8201247234400049,
    "legacy_fitness": 0.8201247234400049,
    "target_cdf": 0.9997673708708951,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.84755802154541,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8003210966298503,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.84755802154541,
        "normalized": 0.9997673708708951,
        "contribution": 0.49988368543544753
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1110897064208984,
        "normalized": 0.25457682692833283,
        "contribution": 0.06364420673208321
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.824615478515625,
        "normalized": 0.9759544616328187,
        "contribution": 0.1463931692449228
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9064433574676514,
        "normalized": 0.9040003521739675,
        "contribution": 0.09040003521739676
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d0e97a8ed5b3a6411479c5748b3eb39512a0c16dba7868a0be974d0900605ad3",
    "id": "ad1d3a3881d6b0f7f9fc7314c63e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": 0.6906055212020874,
    "voiceclap": -0.06378492712974548,
    "fitness": 0.7688799738142611,
    "legacy_fitness": 0.7688799738142611,
    "target_cdf": 0.8542158920076929,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3071789741516113,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6361324362824341,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3071789741516113,
        "normalized": 0.8542158920076929,
        "contribution": 0.42710794600384644
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8091652393341064,
        "normalized": 0.1567722644593298,
        "contribution": 0.03919306611483245
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1522183418273926,
        "normalized": 0.6942076233845773,
        "contribution": 0.1041311435076866
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8921267986297607,
        "normalized": 0.6570028065606865,
        "contribution": 0.06570028065606866
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "71270d6ae800a7cf605f06151109f7616928354cf5706c17813a5711d211e14c",
    "id": "ad41837a9d4a45d8dda5d797b39a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 3.4937667846679688,
    "wer": 0.0,
    "identity": 0.5576834082603455,
    "voiceclap": 0.17985737323760986,
    "fitness": 0.7932672016909528,
    "legacy_fitness": 0.7932672016909528,
    "target_cdf": 0.9811510787587977,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3704047203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7912948436056214,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3704047203063965,
        "normalized": 0.9811510787587977,
        "contribution": 0.49057553937939885
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2128617763519287,
        "normalized": 0.29357562439950946,
        "contribution": 0.07339390609987737
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.335350036621094,
        "normalized": 0.93905329606414,
        "contribution": 0.140857994409621
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.677617073059082,
        "normalized": 0.8646740371672423,
        "contribution": 0.08646740371672423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4523831ca441a917406ac75cb142379ae525d7e07a9e9a09e5f7b9eaad622985",
    "id": "ad697314207a739b606b6411edbf",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 3.773050308227539,
    "wer": 0.0,
    "identity": 0.7606734037399292,
    "voiceclap": -0.024051178246736526,
    "fitness": 0.7895127461433411,
    "legacy_fitness": 0.7895127461433411,
    "target_cdf": 0.9503685407002768,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9240121841430664,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7268292372582942,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9240121841430664,
        "normalized": 0.9503685407002768,
        "contribution": 0.4751842703501384
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.917250871658325,
        "normalized": 0.8809377370125135,
        "contribution": 0.13214066055187704
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.378382682800293,
        "normalized": 0.7984049702519815,
        "contribution": 0.07984049702519816
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "479e502e2b560b898edaae2c31fb335fff70d931daa6233f72d1f86c2898c9d3",
    "id": "ad715b1d410c0f70663426805e4c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.55477237701416,
    "genuineness": 4.3504462242126465,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.2544487714767456,
    "fitness": 0.8052254242952012,
    "legacy_fitness": 0.8052254242952012,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6571283200818057,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5784813165664673,
        "normalized": 0.10143141696818753,
        "contribution": 0.02535785424204688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3301327228546143,
        "normalized": 0.7466748313867635,
        "contribution": 0.11200122470801452
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.712806224822998,
        "normalized": 0.59681718937424,
        "contribution": 0.059681718937424
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b31b89e31d500277dac1881dbc13bddb4884476a2b7e604a17ac72a5a3a254fe",
    "id": "ad76cf5af11e69414cede07b0f64",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 3.600459337234497,
    "wer": 0.0,
    "identity": 0.7505375146865845,
    "voiceclap": 0.1223302036523819,
    "fitness": 0.7808762971249088,
    "legacy_fitness": 0.7808762971249088,
    "target_cdf": 0.9470810897522896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.891547203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7389574023931897,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.891547203063965,
        "normalized": 0.9470810897522896,
        "contribution": 0.4735405448761448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0296716690063477,
        "normalized": 0.2254527264287831,
        "contribution": 0.05636318160719578
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.774919033050537,
        "normalized": 0.8542158295589685,
        "contribution": 0.12813237443384526
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4223875999450684,
        "normalized": 0.8092130147600393,
        "contribution": 0.08092130147600393
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c1e58a86d5a68f53dc34610e2e19fb4eced2e51d72a32f3baa68cfc9e5a29838",
    "id": "adaac374eaf3b272592a2b08e34c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 3.6663577556610107,
    "wer": 0.0,
    "identity": 0.5340398550033569,
    "voiceclap": 0.08151784539222717,
    "fitness": 0.7976374958847077,
    "legacy_fitness": 0.7976374958847077,
    "target_cdf": 0.9755090862302205,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.256777286529541,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7942685549790015,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.256777286529541,
        "normalized": 0.9755090862302205,
        "contribution": 0.48775454311511024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1755452156066895,
        "normalized": 0.27895870121176836,
        "contribution": 0.06973967530294209
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.637804985046387,
        "normalized": 0.965045519092528,
        "contribution": 0.1447568278638792
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.020856857299805,
        "normalized": 0.9201750869706999,
        "contribution": 0.09201750869707
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "616950574c44955011e7b4dfb2ad9f21fc2030ed6ee8c37357e650f2f3d2e2c9",
    "id": "ae65c1661541d825dfcda9e1b8bb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12329889088869095,
    "fitness": 0.8149928035477078,
    "legacy_fitness": 0.8149928035477078,
    "target_cdf": 0.9716911200959037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.191847801208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7153990204070804,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.191847801208496,
        "normalized": 0.9716911200959037,
        "contribution": 0.4858455600479519
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.598835825920105,
        "normalized": 0.10564978306864803,
        "contribution": 0.026412445767162007
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.979520797729492,
        "normalized": 0.8914683718087433,
        "contribution": 0.1337202557713115
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.008740186691284,
        "normalized": 0.6942075882065499,
        "contribution": 0.069420758820655
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e4644673f471040909f0fcf443c40a72652e0a623fe36e2c9d2e814008335182",
    "id": "aeb113852835f70dfe4bc33f2638",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.027577362954616547,
    "fitness": 0.8403878865763392,
    "legacy_fitness": 0.8403878865763392,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7853058254258982,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.982177972793579,
        "normalized": 0.2093744108235418,
        "contribution": 0.05234360270588545
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.548847675323486,
        "normalized": 0.9585742212718315,
        "contribution": 0.1437861331907747
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.020856857299805,
        "normalized": 0.9201750869706999,
        "contribution": 0.09201750869707
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bbb0f6b5c66649bcec6568ac0538994b228249bd78d72b9b60bb33483ca632bd",
    "id": "aeb771f1dd368bb87dc6fc5202ca",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.124508857727051,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04988662153482437,
    "fitness": 0.8065336992957364,
    "legacy_fitness": 0.8065336992957364,
    "target_cdf": 0.9495623609658421,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.915895938873291,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.695362991827516,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.915895938873291,
        "normalized": 0.9495623609658421,
        "contribution": 0.47478118048292106
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6802536249160767,
        "normalized": 0.12378946157536314,
        "contribution": 0.030947365393840784
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6414833068847656,
        "normalized": 0.8257492948479319,
        "contribution": 0.12386239422718978
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.89432692527771,
        "normalized": 0.6577205172356444,
        "contribution": 0.06577205172356444
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fda5aebca665bdee7204c98272daf16bbb2b0186d1c4a5802d9dbc87b89dfbb2",
    "id": "af0edcc6edcf8391a4567a14b7e1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": 0.7607844471931458,
    "voiceclap": 0.049565378576517105,
    "fitness": 0.7922007137961102,
    "legacy_fitness": 0.7922007137961102,
    "target_cdf": 0.946232810502631,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8834309577941895,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7075529744876681,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8834309577941895,
        "normalized": 0.946232810502631,
        "contribution": 0.4731164052513155
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7209625244140625,
        "normalized": 0.13363422338747727,
        "contribution": 0.03340855584686932
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.810502052307129,
        "normalized": 0.8612466639446046,
        "contribution": 0.1291869995916907
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.087949275970459,
        "normalized": 0.7184101379779255,
        "contribution": 0.07184101379779255
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c2571cfccd3972678ecef9e0d4a89790382d0876f9a1a95a4b5dddb46afad46c",
    "id": "af163df621764913e328e04ed12f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 3.973883628845215,
    "wer": 0.06666666666666667,
    "identity": 0.2904157042503357,
    "voiceclap": 0.06847229599952698,
    "fitness": 0.7637154712527848,
    "legacy_fitness": 0.7637154712527848,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8254230044791993,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.334988594055176,
        "normalized": 0.3437155282365555,
        "contribution": 0.08592888205913887
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.922468185424805,
        "normalized": 0.9804199199769799,
        "contribution": 0.14706298799654696
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.320091247558594,
        "normalized": 0.9527255359134541,
        "contribution": 0.09527255359134541
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "894973a3a4902a70cca4b20b2eb51cc318e9f34dba3ffd3bee1264e20f68899a",
    "id": "afada31505208b40977e88ec89e0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 3.835810661315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.037300627678632736,
    "fitness": 0.7964627116935452,
    "legacy_fitness": 0.7964627116935452,
    "target_cdf": 0.947918719056646,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8996634483337402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.73123936456424,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8996634483337402,
        "normalized": 0.947918719056646,
        "contribution": 0.473959359528323
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9448615312576294,
        "normalized": 0.19722848067005766,
        "contribution": 0.049307120167514415
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.952833652496338,
        "normalized": 0.8870402984992682,
        "contribution": 0.13305604477489022
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.193561553955078,
        "normalized": 0.7491684009351236,
        "contribution": 0.07491684009351236
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7304bf6a4d37121e7c185262516cc2a60f01b2b5ef6ce5edb51ebdee588ca1f7",
    "id": "afe28ec6b2c6c74fcacc64b35dc4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.16070696711540222,
    "fitness": 0.7966015824121957,
    "legacy_fitness": 0.7966015824121957,
    "target_cdf": 0.927983714827443,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7292227745056152,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6947656525204122,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7292227745056152,
        "normalized": 0.927983714827443,
        "contribution": 0.4639918574137215
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7752411365509033,
        "normalized": 0.1475786358230735,
        "contribution": 0.03689465895576838
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.61924409866333,
        "normalized": 0.8206824197073732,
        "contribution": 0.12310236295610598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0527453422546387,
        "normalized": 0.7077677319481637,
        "contribution": 0.07077677319481637
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "23430f237416e25807dd99ef50c82c88976903701d219208f9d5c43931306057",
    "id": "b0134b213edf2dbd8a416048c8f7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.400654315948486,
    "wer": 0.0,
    "identity": 0.6660085320472717,
    "voiceclap": -0.06784974038600922,
    "fitness": 0.8357663916289593,
    "legacy_fitness": 0.8357663916289593,
    "target_cdf": 0.9862070255954634,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5002641677856445,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7168843106892603,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5002641677856445,
        "normalized": 0.9862070255954634,
        "contribution": 0.4931035127977317
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7209625244140625,
        "normalized": 0.13363422338747727,
        "contribution": 0.03340855584686932
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7215447425842285,
        "normalized": 0.8432277384318043,
        "contribution": 0.12648416076477065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.837120532989502,
        "normalized": 0.6388808127988868,
        "contribution": 0.06388808127988868
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b14ebb14d324eb794af9b424eeee73c2c55f2e67f832558bc184dd634403d526",
    "id": "b05b1c9bf5ac675df25975c47fb2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": 0.7706494331359863,
    "voiceclap": -0.1218404769897461,
    "fitness": 0.7253935546254955,
    "legacy_fitness": 0.7253935546254955,
    "target_cdf": 0.7894999922465963,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.047459602355957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5882600039658876,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.047459602355957,
        "normalized": 0.7894999922465963,
        "contribution": 0.39474999612329814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8396968841552734,
        "normalized": 0.16536210020893177,
        "contribution": 0.04134052505223294
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.932048797607422,
        "normalized": 0.6234115059757088,
        "contribution": 0.09351172589635633
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.683102607727051,
        "normalized": 0.5865775689400016,
        "contribution": 0.05865775689400016
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e42442049a3318c9a6156efea9cab46aaf6fe529c601bfdd687d4a93c2c8a714",
    "id": "b0e21228de76b4c214bc83438698",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.055472373962402,
    "wer": 0.03571428571428571,
    "identity": 0.6592797040939331,
    "voiceclap": 0.06336652487516403,
    "fitness": 0.7939210558597865,
    "legacy_fitness": 0.7939210558597865,
    "target_cdf": 0.9952532227437393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9060754776000977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8619299117673718,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9060754776000977,
        "normalized": 0.9952532227437393,
        "contribution": 0.49762661137186964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.6526029109954834,
        "normalized": 0.4853935580062482,
        "contribution": 0.12134838950156206
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.780136585235596,
        "normalized": 0.9736578749090112,
        "contribution": 0.14604868123635167
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.540116310119629,
        "normalized": 0.9690622965758837,
        "contribution": 0.09690622965758838
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1606e2ae1946e9f40b1f72f6d3f5ba3c03a7d1f791da4758ff8405512d030de2",
    "id": "b131d286627acd2e19b3dd64b3ee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03294993191957474,
    "fitness": 0.7884644206648161,
    "legacy_fitness": 0.7884644206648161,
    "target_cdf": 0.9427315386702796,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.850965976715088,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7737605749613461,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.850965976715088,
        "normalized": 0.9427315386702796,
        "contribution": 0.4713657693351398
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2976720333099365,
        "normalized": 0.3280469413306799,
        "contribution": 0.08201173533266998
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.872772216796875,
        "normalized": 0.8729869940654369,
        "contribution": 0.13094804910981553
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8448362350463867,
        "normalized": 0.8943502118372078,
        "contribution": 0.08943502118372078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6160542fdbdf277762c99363c83659833f372bc92845e3af6fc2791bf0d6adf8",
    "id": "b230664f51442ee501fb4f927d43",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.149696350097656,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": 0.3671860098838806,
    "voiceclap": -0.16295883059501648,
    "fitness": 0.7402280254132496,
    "legacy_fitness": 0.7402280254132496,
    "target_cdf": 0.7984049961898617,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.0799245834350586,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6236323808194085,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.0799245834350586,
        "normalized": 0.7984049961898617,
        "contribution": 0.39920249809493086
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.131443977355957,
        "normalized": 0.2621540050393544,
        "contribution": 0.0655385012598386
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.9342727661132812,
        "normalized": 0.6241528877228036,
        "contribution": 0.09362293315842055
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.878925323486328,
        "normalized": 0.6526844830621856,
        "contribution": 0.06526844830621857
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ad1285b1b1fbfed4577e36932b4f51c9319b03ebf63153a4d5574373e1c58d55",
    "id": "b24dbb6eb7730085a42bf2f798f7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.20609712600708,
    "wer": 0.0,
    "identity": 0.6175183057785034,
    "voiceclap": -0.1489296406507492,
    "fitness": 0.8241390340617109,
    "legacy_fitness": 0.8241390340617109,
    "target_cdf": 0.9811510787587977,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3704047203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7623282185545692,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3704047203063965,
        "normalized": 0.9811510787587977,
        "contribution": 0.49057553937939885
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0568110942840576,
        "normalized": 0.23494557063169752,
        "contribution": 0.05873639265792438
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.872772216796875,
        "normalized": 0.8729869940654369,
        "contribution": 0.13094804910981553
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4707932472229004,
        "normalized": 0.8206823740743052,
        "contribution": 0.08206823740743052
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8bad755845a95333e079328be626d5b8b1b192937c015d605af05582bb396b9e",
    "id": "b25df5c7a61d9f7faf7caa9002fe",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 3.886019229888916,
    "wer": 0.0,
    "identity": 0.48953306674957275,
    "voiceclap": 0.0293625146150589,
    "fitness": 0.7919355686842685,
    "legacy_fitness": 0.7919355686842685,
    "target_cdf": 0.9418287707049221,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8428497314453125,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7279613784229597,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8428497314453125,
        "normalized": 0.9418287707049221,
        "contribution": 0.47091438535246105
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9855704307556152,
        "normalized": 0.2105000137670302,
        "contribution": 0.05262500344175755
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7838149070739746,
        "normalized": 0.8559956150601711,
        "contribution": 0.12839934225902566
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.233165979385376,
        "normalized": 0.760226473697154,
        "contribution": 0.07602264736971541
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "03ffa562a9f3ad14b573d9d0a675b8d1d2425357f7f90b04812c7c49960db5a7",
    "id": "b27528a243281ae627ce115f7f73",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.9676074981689453,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03736254572868347,
    "fitness": 0.7975047275031619,
    "legacy_fitness": 0.7975047275031619,
    "target_cdf": 0.9453737983316208,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.875314712524414,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6735596597228637,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.875314712524414,
        "normalized": 0.9453737983316208,
        "contribution": 0.4726868991658104
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5525259971618652,
        "normalized": 0.8049331728558216,
        "contribution": 0.12073997592837322
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6478986740112305,
        "normalized": 0.574365695933482,
        "contribution": 0.0574365695933482
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f9df360e0fac3bb759e33da9e1707ece3e9acbe643ec429a51635962263591a3",
    "id": "b304372ab113aa3617e4e052c39b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 4.7834930419921875,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": -0.10655046254396439,
    "fitness": 0.7052501347591297,
    "legacy_fitness": 0.7052501347591297,
    "target_cdf": 0.746674830940225,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.901367425918579,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5351706395073829,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.901367425918579,
        "normalized": 0.746674830940225,
        "contribution": 0.3733374154701125
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9109374284744263,
        "normalized": 0.18656518874247485,
        "contribution": 0.04664129718561871
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.508389949798584,
        "normalized": 0.47702682157139203,
        "contribution": 0.07155402323570881
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.256253719329834,
        "normalized": 0.4363790361594285,
        "contribution": 0.04363790361594286
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "98872e1c664e4a8170b7df28fbb7001488878f8ad52b4081d9aee7ba87d75aba",
    "id": "b3084d2bd1ef95be9518f9640f0d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04157485067844391,
    "fitness": 0.7630233176047585,
    "legacy_fitness": 0.7630233176047585,
    "target_cdf": 0.8680429235090776,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3721086978912354,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6874054977827833,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3721086978912354,
        "normalized": 0.8680429235090776,
        "contribution": 0.4340214617545388
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0296716690063477,
        "normalized": 0.2254527264287831,
        "contribution": 0.05636318160719578
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.463568687438965,
        "normalized": 0.7826722604742027,
        "contribution": 0.1174008390711304
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.369581699371338,
        "normalized": 0.7962001534991823,
        "contribution": 0.07962001534991824
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ca4a5b2fcc31c2a96b10bfb7b1fb238a55fdd967e4b76a9b35802de14941660a",
    "id": "b325c27ed26296b6863d4e8cf2d9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.44,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.021592434495687485,
    "fitness": 0.8112430106268589,
    "legacy_fitness": 0.8112430106268589,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7296998824660885,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.632759928703308,
        "normalized": 0.11295969157872005,
        "contribution": 0.02823992289468001
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.997312545776367,
        "normalized": 0.8943502454552662,
        "contribution": 0.13415253681828992
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3167755603790283,
        "normalized": 0.782672257515953,
        "contribution": 0.07826722575159531
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d130aff974f4877c50ec31bdfce2344f60d6ccfebb2a82bb6d79201ced8f3618",
    "id": "b38c988b3f3d34299e8830ef8ff9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.7165660858154297,
    "wer": 0.0,
    "identity": 0.553689181804657,
    "voiceclap": 0.050379425287246704,
    "fitness": 0.7873518477497667,
    "legacy_fitness": 0.7873518477497667,
    "target_cdf": 0.9495623609658421,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.915895938873291,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7007822650936437,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.915895938873291,
        "normalized": 0.9495623609658421,
        "contribution": 0.47478118048292106
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6191902160644531,
        "normalized": 0.1099935309435226,
        "contribution": 0.02749838273588065
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.828293561935425,
        "normalized": 0.8646740370257361,
        "contribution": 0.1297011055538604
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9889378547668457,
        "normalized": 0.6880159632098157,
        "contribution": 0.06880159632098158
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "40d40a5287f17d16f8c2b4df4cee629ea38682fd741d730fd3fa18b93c24f833",
    "id": "b3979e0eb143cd5660ff6fb7d26a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.381826400756836,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.17135834693908691,
    "fitness": 0.8074819583088888,
    "legacy_fitness": 0.8074819583088888,
    "target_cdf": 0.9331928210240071,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.769804000854492,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6890880200126637,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.769804000854492,
        "normalized": 0.9331928210240071,
        "contribution": 0.46659641051200357
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4813601970672607,
        "normalized": 0.7872381937406205,
        "contribution": 0.11808572906109308
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0043396949768066,
        "normalized": 0.692836384904358,
        "contribution": 0.0692836384904358
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b5b045d4ebf228303effb022ed8376b2925bb8b5355be1d9cfe7fcbc6d6db451",
    "id": "b39a840ba56658cc5f1650bfe9f5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": 0.7283890247344971,
    "voiceclap": 0.07054451107978821,
    "fitness": 0.8138491485263064,
    "legacy_fitness": 0.8138491485263064,
    "target_cdf": 0.9996706910421757,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.750163555145264,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8200700219919373,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.750163555145264,
        "normalized": 0.9996706910421757,
        "contribution": 0.4998353455210878
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2332162857055664,
        "normalized": 0.3016951493925215,
        "contribution": 0.07542378734813038
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.278297424316406,
        "normalized": 0.9912255261486401,
        "contribution": 0.148683828922296
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.425703048706055,
        "normalized": 0.9612706020042315,
        "contribution": 0.09612706020042316
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "07d53363d6f7172a5a100fadd6cd573a034541fb4866355891a3100bc2452fc0",
    "id": "b3e5023880134209ce884831c57c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.362998008728027,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.030993377789855003,
    "fitness": 0.8378493896952081,
    "legacy_fitness": 0.8378493896952081,
    "target_cdf": 0.9935110473296658,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7924485206604004,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7506196377983969,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7924485206604004,
        "normalized": 0.9935110473296658,
        "contribution": 0.4967555236648329
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.826127290725708,
        "normalized": 0.16150748713874752,
        "contribution": 0.04037687178468688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.068478107452393,
        "normalized": 0.9053242349151492,
        "contribution": 0.13579863523727237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2947731018066406,
        "normalized": 0.7768860711160478,
        "contribution": 0.07768860711160479
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bf76f39e57c1900e27dec6c657113d01226402201376b066bf38af83dcc5cebe",
    "id": "b3e57d3f48820275b9c9fe6367d6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.3504462242126465,
    "wer": 0.0,
    "identity": 0.5937182903289795,
    "voiceclap": -0.07214721292257309,
    "fitness": 0.8372938918210997,
    "legacy_fitness": 0.8372938918210997,
    "target_cdf": 0.9970202380692827,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.0684003829956055,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7402617246716551,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.0684003829956055,
        "normalized": 0.9970202380692827,
        "contribution": 0.4985101190346414
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5038483142852783,
        "normalized": 0.08701389328623654,
        "contribution": 0.021753473321559136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.362037181854248,
        "normalized": 0.9418287736823874,
        "contribution": 0.1412743160523581
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3343775272369385,
        "normalized": 0.787238162630965,
        "contribution": 0.0787238162630965
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a7589dcec7d4c1bbf5ae92c3ceea5a030b27719badf0b6f823f5bee31eee7a03",
    "id": "b41af7d5c45d7c244cb6e6e16ff9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.074300289154053,
    "wer": 0.0,
    "identity": 0.6585050225257874,
    "voiceclap": -0.06560111790895462,
    "fitness": 0.7716971317615802,
    "legacy_fitness": 0.7716971317615802,
    "target_cdf": 0.8824847618116329,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4451546669006348,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6469355703664579,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4451546669006348,
        "normalized": 0.8824847618116329,
        "contribution": 0.44124238090581647
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5309876203536987,
        "normalized": 0.09206841268091792,
        "contribution": 0.02301710317022948
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.521390914916992,
        "normalized": 0.7973043411623713,
        "contribution": 0.11959565117435568
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.812917709350586,
        "normalized": 0.6308043511605621,
        "contribution": 0.06308043511605621
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fb888095dd84d69977cbd5d621b607b5daa74cca05ae6b6aebb37f227681207a",
    "id": "b41da2fd587746e05cd458272f59",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.243753433227539,
    "wer": 0.03571428571428571,
    "identity": 0.5261814594268799,
    "voiceclap": 0.01031503640115261,
    "fitness": 0.8005543908124698,
    "legacy_fitness": 0.8005543908124698,
    "target_cdf": 0.9972925007937494,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.100865364074707,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7369614081736097,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.100865364074707,
        "normalized": 0.9972925007937494,
        "contribution": 0.4986462503968747
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5174179077148438,
        "normalized": 0.0895146120757892,
        "contribution": 0.0223786530189473
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.26418399810791,
        "normalized": 0.9311452345235685,
        "contribution": 0.13967178517853526
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.241966962814331,
        "normalized": 0.762647195792524,
        "contribution": 0.0762647195792524
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0ca5843c241892faaf7b6b9756005b76efd00ea2cf12ac8049e86e9d49abe3b2",
    "id": "b493b483660b6eefed50ad44188b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 4.375550270080566,
    "wer": 0.0,
    "identity": 0.5955134034156799,
    "voiceclap": -0.15110844373703003,
    "fitness": 0.7622382138220076,
    "legacy_fitness": 0.7622382138220076,
    "target_cdf": 0.8487882165208668,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.282830238342285,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.719315148741629,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.282830238342285,
        "normalized": 0.8487882165208668,
        "contribution": 0.4243941082604334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5258116722106934,
        "normalized": 0.42754922115757454,
        "contribution": 0.10688730528939364
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1878011226654053,
        "normalized": 0.7050781385006836,
        "contribution": 0.10576172077510253
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4795942306518555,
        "normalized": 0.8227201441669949,
        "contribution": 0.0822720144166995
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b1d7f3d06d64d0673b46e488b3ecb23acb419f04fd89d1663ffcbf00ce1bf498",
    "id": "b4e834a13b1586840ed5696cf26e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.483288764953613,
    "genuineness": 4.193545341491699,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.02647845819592476,
    "fitness": 0.7550995727908033,
    "legacy_fitness": 0.7550995727908033,
    "target_cdf": 0.836572595542951,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.230074644088745,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6114845998406304,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.230074644088745,
        "normalized": 0.836572595542951,
        "contribution": 0.4182862977714755
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8702285289764404,
        "normalized": 0.17425071298327421,
        "contribution": 0.043562678245818554
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.032125949859619,
        "normalized": 0.6562844989859683,
        "contribution": 0.09844267484789525
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4703657627105713,
        "normalized": 0.511929489754411,
        "contribution": 0.051192948975441105
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6e004486d6262762c90d033f7c4e9fb1a717116a6d464aefcce95643d99d3a62",
    "id": "b5d55d9f183031fc0ecda5ef73ad",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.1998209953308105,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10163326561450958,
    "fitness": 0.8233461550614358,
    "legacy_fitness": 0.8233461550614358,
    "target_cdf": 0.9741313242397945,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.232429027557373,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6821343104459516,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.232429027557373,
        "normalized": 0.9741313242397945,
        "contribution": 0.48706566211989727
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.3545823097229004,
        "normalized": 0.0628538085933158,
        "contribution": 0.01571345214832895
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8016064167022705,
        "normalized": 0.8595110162463788,
        "contribution": 0.1289266524369568
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4487757682800293,
        "normalized": 0.5042854374076865,
        "contribution": 0.050428543740768654
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "58a358f83cc3deb605f94ae7f549c681f0a676dba066d9ffe98a3aaa61dc4d88",
    "id": "b61e79a827b6f0ca2e07c7cefecc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.017816066741943,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.01723785698413849,
    "fitness": 0.8240923654463197,
    "legacy_fitness": 0.8240923654463197,
    "target_cdf": 0.997542097454529,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.13332986831665,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8027624317971778,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.13332986831665,
        "normalized": 0.997542097454529,
        "contribution": 0.4987710487272645
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.124659299850464,
        "normalized": 0.2596155287980314,
        "contribution": 0.06490388219950785
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.691179275512695,
        "normalized": 0.9685129990310137,
        "contribution": 0.14527694985465206
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.170474052429199,
        "normalized": 0.9381055101575337,
        "contribution": 0.09381055101575338
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "56041eb115363e3518cec805c80935849e5df04c5a274fce2a425d52672c576b",
    "id": "b62f99c1403218429956bb7e9bad",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.181467056274414,
    "genuineness": 4.413206577301025,
    "wer": 0.0,
    "identity": 0.5110782384872437,
    "voiceclap": -0.03366018459200859,
    "fitness": 0.8079469388690013,
    "legacy_fitness": 0.8079469388690013,
    "target_cdf": 0.9390533101491965,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8185012340545654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7044224183382057,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8185012340545654,
        "normalized": 0.9390533101491965,
        "contribution": 0.46952665507459823
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.012709617614746,
        "normalized": 0.2196317185230387,
        "contribution": 0.054907929630759675
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.383507251739502,
        "normalized": 0.7614385460618895,
        "contribution": 0.11421578190928343
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.89432692527771,
        "normalized": 0.6577205172356444,
        "contribution": 0.06577205172356444
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "419f08446f11750dcb9d8fe3d16c8d9eaa986b1f4b8143b265cd37c961b1ae4d",
    "id": "b63976e50093f1e36ba83d446182",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 4.21237325668335,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06819455325603485,
    "fitness": 0.8007876118510566,
    "legacy_fitness": 0.8007876118510566,
    "target_cdf": 0.9311452866573339,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7535715103149414,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6908244639738247,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7535715103149414,
        "normalized": 0.9311452866573339,
        "contribution": 0.46557264332866694
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6459312438964844,
        "normalized": 0.826751667074274,
        "contribution": 0.1240127500611411
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.997738838195801,
        "normalized": 0.6907744818671719,
        "contribution": 0.0690774481867172
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8eb06fe003ca34c6c0e7c14e0abd889654018c7691394ea96d45e87f14809b93",
    "id": "b68d94a8e1405d3bf352a54293d2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.036666177213191986,
    "fitness": 0.8276841259126095,
    "legacy_fitness": 0.8276841259126095,
    "target_cdf": 0.9937903356798943,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.808681011199951,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7384732817281956,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.808681011199951,
        "normalized": 0.9937903356798943,
        "contribution": 0.49689516783994714
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.335350036621094,
        "normalized": 0.93905329606414,
        "contribution": 0.140857994409621
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.096750259399414,
        "normalized": 0.721041236201374,
        "contribution": 0.07210412362013739
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "afcd41b91026f80cd59c06bcccdbea09090177c6e8a317b41d93089d4523eab3",
    "id": "b6d8b4628705f8259275e68ae55b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 3.6663577556610107,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.11343811452388763,
    "fitness": 0.7882107724762957,
    "legacy_fitness": 0.7882107724762957,
    "target_cdf": 0.947918719056646,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8996634483337402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7905594401187701,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8996634483337402,
        "normalized": 0.947918719056646,
        "contribution": 0.473959359528323
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5113937854766846,
        "normalized": 0.42104577256937253,
        "contribution": 0.10526144314234313
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7037534713745117,
        "normalized": 0.8394469854437661,
        "contribution": 0.1259170478165649
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6248111724853516,
        "normalized": 0.8542158963153907,
        "contribution": 0.08542158963153908
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a01c2224596513ac060288f70f11693575f6f44178a95881b48289724b6759bc",
    "id": "b6f8b6756cbf462f8fd0d9b7c29e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": 0.6289687156677246,
    "voiceclap": 0.02641548216342926,
    "fitness": 0.8091348772268976,
    "legacy_fitness": 0.8091348772268976,
    "target_cdf": 0.9619218677325021,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0538716316223145,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.707296402253348,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0538716316223145,
        "normalized": 0.9619218677325021,
        "contribution": 0.48096093386625105
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6056206226348877,
        "normalized": 0.10708370011583324,
        "contribution": 0.02677092502895831
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7571277618408203,
        "normalized": 0.8506121673397089,
        "contribution": 0.12759182510095632
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0923497676849365,
        "normalized": 0.719727182571823,
        "contribution": 0.07197271825718231
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4b8f2f674d23c1f207521ff8016b755437ffac745f119ca404808580c4253329",
    "id": "b71ec05193287acabb2aa7000ef8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 5.95112943649292,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06096745282411575,
    "fitness": 0.7942492501559782,
    "legacy_fitness": 0.7942492501559782,
    "target_cdf": 0.9571700447048601,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.997058153152466,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8328413828238728,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.997058153152466,
        "normalized": 0.9571700447048601,
        "contribution": 0.47858502235243006
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.590691566467285,
        "normalized": 0.4570335333180516,
        "contribution": 0.1142583833295129
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.620013236999512,
        "normalized": 0.9638222528181529,
        "contribution": 0.14457333792272292
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.337693214416504,
        "normalized": 0.9542463921920685,
        "contribution": 0.09542463921920685
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "11e2b7e3999539cc25c65ec8aa30d9753c760a7e026f445abfb81070e0892c28",
    "id": "b71fd72e2ca5a133ccf852da7aee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.18726921081543,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11215794086456299,
    "fitness": 0.799408799022387,
    "legacy_fitness": 0.799408799022387,
    "target_cdf": 0.9269054842192939,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.72110652923584,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6953902687141645,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.72110652923584,
        "normalized": 0.9269054842192939,
        "contribution": 0.46345274210964693
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.66817045211792,
        "normalized": 0.8317082004701283,
        "contribution": 0.12475623007051924
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9779367446899414,
        "normalized": 0.6845529821840911,
        "contribution": 0.06845529821840911
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "44e7e2fea8229cd2248ee047a1147e0819afa3fba587a008bf6d5fff603278d4",
    "id": "b726e75730391896b5f09a3309f8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.268857479095459,
    "wer": 0.06666666666666667,
    "identity": 0.5200984477996826,
    "voiceclap": -0.12576048076152802,
    "fitness": 0.7377637247005652,
    "legacy_fitness": 0.7377637247005652,
    "target_cdf": 0.911742319044517,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6155953407287598,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7363674355724283,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6155953407287598,
        "normalized": 0.911742319044517,
        "contribution": 0.4558711595222585
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3773937225341797,
        "normalized": 0.3618509614458445,
        "contribution": 0.09046274036146112
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2322797775268555,
        "normalized": 0.7184101418133952,
        "contribution": 0.10776152127200929
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4795942306518555,
        "normalized": 0.8227201441669949,
        "contribution": 0.0822720144166995
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7e78320baf75fe710f0421679b2d4f9f5e4f7fb7ccc64cdee9ce2281e4fe1b29",
    "id": "b73087d8ed7c7dda1872c794d504",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.2249250411987305,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.17065708339214325,
    "fitness": 0.8282013072788172,
    "legacy_fitness": 0.8282013072788172,
    "target_cdf": 0.98254389250665,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.402869701385498,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.735028586125161,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.402869701385498,
        "normalized": 0.98254389250665,
        "contribution": 0.491271946253325
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.299173593521118,
        "normalized": 0.7780502760142334,
        "contribution": 0.07780502760142334
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "939f2dfb7ce3dd1fed6159648108fdbbcd26837041f1298d904a62fb36593a52",
    "id": "b75350b0a134dabff81f0dc755f2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1592964380979538,
    "fitness": 0.7720451634533263,
    "legacy_fitness": 0.7720451634533263,
    "target_cdf": 0.8595110498680782,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3315277099609375,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6856382161501711,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3315277099609375,
        "normalized": 0.8595110498680782,
        "contribution": 0.4297555249340391
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2366087436676025,
        "normalized": 0.30305817864738066,
        "contribution": 0.07576454466184517
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2589669227600098,
        "normalized": 0.7262672960496188,
        "contribution": 0.10894009440744282
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0659468173980713,
        "normalized": 0.7117805214684402,
        "contribution": 0.07117805214684401
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9dd644953916404d067c170b74ddb1d4ed98e1c28598da36e044b535d608c570",
    "id": "b772a74ed3fcd98bc55602b1fbf9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.045917630195617676,
    "fitness": 0.8300572601755806,
    "legacy_fitness": 0.8300572601755806,
    "target_cdf": 0.9952532227437393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9060754776000977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7509821595795989,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9060754776000977,
        "normalized": 0.9952532227437393,
        "contribution": 0.49762661137186964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6666839122772217,
        "normalized": 0.12062361006755057,
        "contribution": 0.030155902516887642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.5132646560668945,
        "normalized": 0.9557276577709561,
        "contribution": 0.14335914866564342
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.378382682800293,
        "normalized": 0.7984049702519815,
        "contribution": 0.07984049702519816
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "dda3e29d6f48cbbb3b7904f63666a4a511f68a5748680f2b01f54f3255a713fd",
    "id": "b871cc076f8a7e681f72e10f74dc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.55477237701416,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07484237104654312,
    "fitness": 0.822566529451453,
    "legacy_fitness": 0.822566529451453,
    "target_cdf": 0.9606102950148648,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0376393795013428,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6974827940852614,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0376393795013428,
        "normalized": 0.9606102950148648,
        "contribution": 0.4803051475074324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6666839122772217,
        "normalized": 0.12062361006755057,
        "contribution": 0.030155902516887642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6726183891296387,
        "normalized": 0.8326884740215883,
        "contribution": 0.12490327110323823
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7843143939971924,
        "normalized": 0.6211847295770306,
        "contribution": 0.062118472957703064
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "34235cb57663b7b4b007a4ea5cd339bf49becc92d1b6a87b14911fa0c8d627e9",
    "id": "b8b9a28332fa2c5efb2ac36f9865",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11666711419820786,
    "fitness": 0.7893442826636918,
    "legacy_fitness": 0.7893442826636918,
    "target_cdf": 0.8943502420521354,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.510084629058838,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6285787934638377,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.510084629058838,
        "normalized": 0.8943502420521354,
        "contribution": 0.4471751210260677
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5377724170684814,
        "normalized": 0.09336535255322709,
        "contribution": 0.023341338138306772
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4190900325775146,
        "normalized": 0.7710131225645452,
        "contribution": 0.11565196838468178
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.2210495471954346,
        "normalized": 0.4241036591478142,
        "contribution": 0.04241036591478142
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b451031482ae5d29d80bbfe22ee05e8020e83bdd06c28d208a04999be16f2287",
    "id": "ba2bac5db2af091484135a5d0f6a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 3,
    "generation": null,
    "ce": 6.562715530395508,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": 0.38263416290283203,
    "voiceclap": 0.030862249433994293,
    "fitness": 0.7186862101932148,
    "legacy_fitness": 0.7186862101932148,
    "target_cdf": 0.773372656261877,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.9906460046768188,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.618710806615628,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.9906460046768188,
        "normalized": 0.773372656261877,
        "contribution": 0.3866863281309385
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2552669048309326,
        "normalized": 0.3106034035168039,
        "contribution": 0.07765085087920097
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.7563581466674805,
        "normalized": 0.5636209621748036,
        "contribution": 0.08454314432622054
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.021941661834717,
        "normalized": 0.6983048327926806,
        "contribution": 0.06983048327926807
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "55fccfebb1d08bfc443c2151123e72862105f0b122bf1de4e5be370f5801f78c",
    "id": "ba5fe76e2f2a1765e24a821f6686",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": 0.7182809114456177,
    "voiceclap": 0.02146226540207863,
    "fitness": 0.7964059052274568,
    "legacy_fitness": 0.7964059052274568,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6888061596881281,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4190900325775146,
        "normalized": 0.7710131225645452,
        "contribution": 0.11565196838468178
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1495563983917236,
        "normalized": 0.7365719667634352,
        "contribution": 0.07365719667634352
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9902ae529db301037ca139aee0db7db35dd3b6d41d203fef8dcabf26ce0e057a",
    "id": "bac03d176ed7b731c880b1e27353",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 3.835810661315918,
    "wer": 0.0,
    "identity": 0.5880889892578125,
    "voiceclap": 0.07144130766391754,
    "fitness": 0.7968598455207547,
    "legacy_fitness": 0.7968598455207547,
    "target_cdf": 0.947918719056646,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8996634483337402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7005639489648178,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8996634483337402,
        "normalized": 0.947918719056646,
        "contribution": 0.473959359528323
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6666839122772217,
        "normalized": 0.12062361006755057,
        "contribution": 0.030155902516887642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.774919033050537,
        "normalized": 0.8542158295589685,
        "contribution": 0.12813237443384526
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9735360145568848,
        "normalized": 0.6831631248576189,
        "contribution": 0.0683163124857619
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "374d637bfdffb6958d4a0cefffd2da4aad76d90fbd0a963cd34fa67081e730a8",
    "id": "badc7bbabdc0324190811a6526f6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 3.961331605911255,
    "wer": 0.0,
    "identity": 0.7017697095870972,
    "voiceclap": 0.09442931413650513,
    "fitness": 0.8191135380549109,
    "legacy_fitness": 0.8191135380549109,
    "target_cdf": 0.9998370237912488,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.944952964782715,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8041847039744004,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.944952964782715,
        "normalized": 0.9998370237912488,
        "contribution": 0.4999185118956244
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0568110942840576,
        "normalized": 0.23494557063169752,
        "contribution": 0.05873639265792438
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.456212043762207,
        "normalized": 0.9943171611150623,
        "contribution": 0.14914757416725935
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.460906982421875,
        "normalized": 0.9638222525359226,
        "contribution": 0.09638222525359226
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a619ec393746e59e18e7ecd3946b09d523d37e250abdc53d22b1926c2fcc668c",
    "id": "bb169ae1d7fd85bb56a86c781070",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07163271307945251,
    "fitness": 0.789922237720896,
    "legacy_fitness": 0.789922237720896,
    "target_cdf": 0.9178304042951448,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6561765670776367,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6792644074279659,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6561765670776367,
        "normalized": 0.9178304042951448,
        "contribution": 0.4589152021475724
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7548866271972656,
        "normalized": 0.14223933958928087,
        "contribution": 0.03555983489732022
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.556973934173584,
        "normalized": 0.8060085940212309,
        "contribution": 0.12090128910318464
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.837120532989502,
        "normalized": 0.6388808127988868,
        "contribution": 0.06388808127988868
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3161ccc732afa1d5f257e1670db067c51ffcfef4bea131a9982c517cdc5617d7",
    "id": "bb8d5a408a7577c4a28c009f6e37",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.00831027515232563,
    "fitness": 0.7897385720100865,
    "legacy_fitness": 0.7897385720100865,
    "target_cdf": 0.9470810897522896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.891547203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7382553399260242,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.891547203063965,
        "normalized": 0.9470810897522896,
        "contribution": 0.4735405448761448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9550387859344482,
        "normalized": 0.20049800895093267,
        "contribution": 0.05012450223773317
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.854980707168579,
        "normalized": 0.8697054717721656,
        "contribution": 0.13045582076582485
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.563203811645508,
        "normalized": 0.8413447204632145,
        "contribution": 0.08413447204632146
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d6254cd4891f6ed3dc2381fa78cbbc74f7d21196948eaef78cec473a4e197a83",
    "id": "bb91d37d23b78e49a5322be9ed9a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": 0.5610711574554443,
    "voiceclap": 0.16756147146224976,
    "fitness": 0.8389340792333551,
    "legacy_fitness": 0.8389340792333551,
    "target_cdf": 0.9997385569881334,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.815093040466309,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7652799844522522,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.815093040466309,
        "normalized": 0.9997385569881334,
        "contribution": 0.4998692784940667
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.768456220626831,
        "normalized": 0.1457841285203917,
        "contribution": 0.03644603213009792
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.7979278564453125,
        "normalized": 0.974597599214758,
        "contribution": 0.1461896398822137
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5015969276428223,
        "normalized": 0.8277503394587389,
        "contribution": 0.08277503394587389
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c440d8a4cd4dca46f50beb63bfcd3a7a2bfc5727010e422e9aba1d7748d3027f",
    "id": "bcab6760199fc314db1cd27dfb54",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2,
    "generation": null,
    "ce": 6.515059471130371,
    "genuineness": 4.18726921081543,
    "wer": 0.0,
    "identity": 0.7508820295333862,
    "voiceclap": 0.09269741922616959,
    "fitness": 0.7644539832327486,
    "legacy_fitness": 0.7644539832327486,
    "target_cdf": 0.8542158920076929,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3071789741516113,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6910670830976096,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3071789741516113,
        "normalized": 0.8542158920076929,
        "contribution": 0.42710794600384644
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.273925304412842,
        "normalized": 0.31822925193686846,
        "contribution": 0.07955731298421712
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.254518985748291,
        "normalized": 0.7249653116605688,
        "contribution": 0.10874479674908531
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2199645042419434,
        "normalized": 0.7565702736046067,
        "contribution": 0.07565702736046068
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f32c07d359eeb01e51601d240537507340a7effe877ef56db038c933c968fa5c",
    "id": "bdc2c9c020fc12daa0ebc006f5d3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.3127153515815735,
    "fitness": 0.7692946095696995,
    "legacy_fitness": 0.7692946095696995,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6062483158514647,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8227348327636719,
        "normalized": 0.16055304509332224,
        "contribution": 0.04013826127333056
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.692976236343384,
        "normalized": 0.5416106285876032,
        "contribution": 0.08124159428814048
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.683102607727051,
        "normalized": 0.5865775689400016,
        "contribution": 0.05865775689400016
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fa9a66d01b8346f0d0a050edc8c893b670730671093eee17c5807e41a9456f38",
    "id": "bddd0fd3dbd998a5f9cbdb66db39",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": 0.6992307305335999,
    "voiceclap": -0.014049151912331581,
    "fitness": 0.7446727216732293,
    "legacy_fitness": 0.7446727216732293,
    "target_cdf": 0.8403976997716803,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.246307134628296,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6351382545916715,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.246307134628296,
        "normalized": 0.8403976997716803,
        "contribution": 0.42019884988584016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8498740196228027,
        "normalized": 0.16829178263767364,
        "contribution": 0.04207294565941841
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2189362049102783,
        "normalized": 0.7144411892689617,
        "contribution": 0.10716617839034426
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8921267986297607,
        "normalized": 0.6570028065606865,
        "contribution": 0.06570028065606866
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d6ee130a35788ba41dd94b42682fcdafa9b85f959ff63a3bcab311aae0d4657e",
    "id": "bde4b8818338cab6c9c233ddb6fa",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.4812145233154297,
    "wer": 0.0,
    "identity": 0.5584417581558228,
    "voiceclap": 0.15630757808685303,
    "fitness": 0.8007889764751766,
    "legacy_fitness": 0.8007889764751766,
    "target_cdf": 0.996049248624995,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.971005439758301,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7825472391834298,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.971005439758301,
        "normalized": 0.996049248624995,
        "contribution": 0.4980246243124975
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9109374284744263,
        "normalized": 0.18656518874247485,
        "contribution": 0.04664129718561871
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.886885643005371,
        "normalized": 0.97888519494114,
        "contribution": 0.146832779241171
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.950448513031006,
        "normalized": 0.9104853844414267,
        "contribution": 0.09104853844414268
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3ac415171e2f940d7344913b867db86c6e534159f5ef166fe321ba5f970c7217",
    "id": "befba9982db84d300337b057f6e5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.141753673553467,
    "genuineness": 4.055472373962402,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.12951532006263733,
    "fitness": 0.7529931341685212,
    "legacy_fitness": 0.7529931341685212,
    "target_cdf": 0.9685130012341955,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1431503295898438,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7601638913699734,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1431503295898438,
        "normalized": 0.9685130012341955,
        "contribution": 0.48425650061709774
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.009317398071289,
        "normalized": 0.21847802587457404,
        "contribution": 0.05461950646864351
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.997312545776367,
        "normalized": 0.8943502454552662,
        "contribution": 0.13415253681828992
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7128210067749023,
        "normalized": 0.8713534746594227,
        "contribution": 0.08713534746594227
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6445db30e0cbe5d649c90628c8590c5e16d5b2a4c03865540f83e5137bfa4fc7",
    "id": "bf050cd5f5411166558c4c4ddb99",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.074300289154053,
    "wer": 0.0,
    "identity": 0.48474258184432983,
    "voiceclap": 0.040106307715177536,
    "fitness": 0.7746004322271283,
    "legacy_fitness": 0.7746004322271283,
    "target_cdf": 0.8914684095185468,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.493852138519287,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6903305483530177,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.493852138519287,
        "normalized": 0.8914684095185468,
        "contribution": 0.4457342047592734
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.033064126968384,
        "normalized": 0.2266273368759334,
        "contribution": 0.05665683421898335
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2678627967834473,
        "normalized": 0.7288620844464377,
        "contribution": 0.10932931266696566
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.329977035522461,
        "normalized": 0.7861019670779525,
        "contribution": 0.07861019670779526
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1de5de666ee6dbb79c9285b1498ed7cb010dab3d0f3190df40764206a067eb6b",
    "id": "bf3c0beea5b5b1d06716c1f05e58",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 3,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.020573200657963753,
    "fitness": 0.8191852092756889,
    "legacy_fitness": 0.8191852092756889,
    "target_cdf": 0.9585742155738795,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0132906436920166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7426496452680766,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0132906436920166,
        "normalized": 0.9585742155738795,
        "contribution": 0.4792871077869397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0262794494628906,
        "normalized": 0.2242816522895154,
        "contribution": 0.05607041307237885
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.281571388244629,
        "normalized": 0.7733726077901995,
        "contribution": 0.07733726077901995
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ebbad71fb8759f50a803a96a3f8bf6e9b154e2468c6e2cc79078b00da46c98bb",
    "id": "bf60687d9e83d20b5f56f424b01c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.1998209953308105,
    "wer": 0.0,
    "identity": 0.3949507176876068,
    "voiceclap": 0.01876463182270527,
    "fitness": 0.8332013305153487,
    "legacy_fitness": 0.8332013305153487,
    "target_cdf": 0.9950331004711068,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.889843463897705,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7892069139579603,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.889843463897705,
        "normalized": 0.9950331004711068,
        "contribution": 0.4975165502355534
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.161975860595703,
        "normalized": 0.27373275044956896,
        "contribution": 0.06843318761239224
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.130748271942139,
        "normalized": 0.9142167170262676,
        "contribution": 0.13713250755394013
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.660015106201172,
        "normalized": 0.861246685560745,
        "contribution": 0.08612466855607451
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a8502fb63df0d0ddecea302511bf745f300db65fb3a22cc2264be2b215e7143c",
    "id": "bf6417d316fd3681808e3cc4eb53",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13720320165157318,
    "fitness": 0.8349334689381435,
    "legacy_fitness": 0.8349334689381435,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7611954570281587,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.002532482147217,
        "normalized": 0.21618097954673832,
        "contribution": 0.05404524488668458
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.997312545776367,
        "normalized": 0.8943502454552662,
        "contribution": 0.13415253681828992
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4707932472229004,
        "normalized": 0.8206823740743052,
        "contribution": 0.08206823740743052
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4b79947ca67ec5dcbbcf394aaced98b30ac69a3bbe0ade0f8c4d668177c8c65f",
    "id": "bf982cfc25a6020a667fdb854fe4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.626256465911865,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13196229934692383,
    "fitness": 0.8270056155854191,
    "legacy_fitness": 0.8270056155854191,
    "target_cdf": 0.9638222590155852,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0782203674316406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6896634401739953,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0782203674316406,
        "normalized": 0.9638222590155852,
        "contribution": 0.4819111295077926
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4631394147872925,
        "normalized": 0.07982495177638921,
        "contribution": 0.019956237944097302
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8460850715637207,
        "normalized": 0.8680429242459118,
        "contribution": 0.13020643863688677
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.652299165725708,
        "normalized": 0.5758963408521858,
        "contribution": 0.05758963408521858
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1ef4f6027a62e1a12f2961fc782efaafd211bfd45cbe80de81c50c3416006e49",
    "id": "bf994ca07138be51d00c6ed24353",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.013416305184364319,
    "fitness": 0.8314247456329995,
    "legacy_fitness": 0.8314247456329995,
    "target_cdf": 0.9970202380692827,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.0684003829956055,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7454355652152209,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.0684003829956055,
        "normalized": 0.9970202380692827,
        "contribution": 0.4985101190346414
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.659899115562439,
        "normalized": 0.11906225606487442,
        "contribution": 0.029765564016218604
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.353141784667969,
        "normalized": 0.9409148986309699,
        "contribution": 0.1411372347946455
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.233165979385376,
        "normalized": 0.760226473697154,
        "contribution": 0.07602264736971541
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "83f9277f34d7abb43992628663409687f63ed25674f0156e9e1ed7c5e38a1241",
    "id": "bf9fe8d37796476d8d51fcdba751",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": 0.6614499688148499,
    "voiceclap": 0.05601932853460312,
    "fitness": 0.8155048933366721,
    "legacy_fitness": 0.8155048933366721,
    "target_cdf": 0.9862070255954634,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5002641677856445,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7345257083137949,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5002641677856445,
        "normalized": 0.9862070255954634,
        "contribution": 0.4931035127977317
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.476709008216858,
        "normalized": 0.08216957897144506,
        "contribution": 0.020542394742861264
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.409186363220215,
        "normalized": 0.806008610702397,
        "contribution": 0.08060086107023971
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "be7fdd65e28ddfb8c29293b75487fdb4f4c4668041f4d9f6dfd1de456ca37e76",
    "id": "bfb89e8817c2d31a48475658f701",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.117925643920898,
    "genuineness": 3.973883628845215,
    "wer": 0.03333333333333333,
    "identity": 0.7096370458602905,
    "voiceclap": 0.04421555995941162,
    "fitness": 0.7857967714277669,
    "legacy_fitness": 0.7857967714277669,
    "target_cdf": 0.9887331723771762,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5814266204833984,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8339490431609855,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5814266204833984,
        "normalized": 0.9887331723771762,
        "contribution": 0.4943665861885881
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5054571628570557,
        "normalized": 0.4183740949029872,
        "contribution": 0.1045935237257468
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.362037181854248,
        "normalized": 0.9418287736823874,
        "contribution": 0.1412743160523581
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.161672592163086,
        "normalized": 0.9371461719429255,
        "contribution": 0.09371461719429255
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f30239ea7fc35ab5fccdca2c17febba966c7b09cf4abe08ea14843ff15cc6687",
    "id": "bfdeaa77d8403842234f56c1727d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": 0.5463110208511353,
    "voiceclap": -0.12110163271427155,
    "fitness": 0.7986603957960282,
    "legacy_fitness": 0.7986603957960282,
    "target_cdf": 0.9247120186784414,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.704874038696289,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7702484662400452,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.704874038696289,
        "normalized": 0.9247120186784414,
        "contribution": 0.4623560093392207
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4588115215301514,
        "normalized": 0.39752227638752075,
        "contribution": 0.09938056909688019
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.685961961746216,
        "normalized": 0.835607112993103,
        "contribution": 0.12534106694896543
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5191988945007324,
        "normalized": 0.8317082085497887,
        "contribution": 0.08317082085497887
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ed399da436cc7997fa01fb42d52748a3eda8d89c6d15e88b311d470457f3686c",
    "id": "c0673ee08e316a4e230747044465",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": 0.728876531124115,
    "voiceclap": -0.013502702116966248,
    "fitness": 0.7690404086586549,
    "legacy_fitness": 0.7690404086586549,
    "target_cdf": 0.8778006373398928,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4208059310913086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6599262227194775,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4208059310913086,
        "normalized": 0.8778006373398928,
        "contribution": 0.4389003186699464
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8736209869384766,
        "normalized": 0.17525676081671665,
        "contribution": 0.04381419020417916
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.30344557762146,
        "normalized": 0.7391168214554646,
        "contribution": 0.11086752321831969
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.911929130554199,
        "normalized": 0.663441906270323,
        "contribution": 0.0663441906270323
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5140d008d8a58c2fb860bbb6f32546b899777c228556bb34548478d3c2894289",
    "id": "c0cfe81c2b23450179d4917b60c6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 4.432034492492676,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04446301609277725,
    "fitness": 0.7484790440842617,
    "legacy_fitness": 0.7484790440842617,
    "target_cdf": 0.8165628585110323,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.1489124298095703,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6490660773439395,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.1489124298095703,
        "normalized": 0.8165628585110323,
        "contribution": 0.40828142925551614
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.307849407196045,
        "normalized": 0.33229177143790456,
        "contribution": 0.08307294285947614
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.7085437774658203,
        "normalized": 0.5470311857899101,
        "contribution": 0.08205467786848651
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2199645042419434,
        "normalized": 0.7565702736046067,
        "contribution": 0.07565702736046068
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "19ab7cf2d062adc63e306ba4e92138b1fa6af181eb3629fddf7f13ca9a5449bb",
    "id": "c145a8577b2d1089cd6cf7eca55a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.157639026641846,
    "genuineness": 4.174716949462891,
    "wer": 0.03333333333333333,
    "identity": 0.3248552083969116,
    "voiceclap": -0.06288018822669983,
    "fitness": 0.7967172038359829,
    "legacy_fitness": 0.7967172038359829,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.772985961558753,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9380767345428467,
        "normalized": 0.19506683634182148,
        "contribution": 0.04876670908545537
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.281975746154785,
        "normalized": 0.9331927987311419,
        "contribution": 0.13997891980967128
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.756826162338257,
        "normalized": 0.8793763702581625,
        "contribution": 0.08793763702581625
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "879a9503215b4212fcd87449995aec806be12db7d5181a2c774c26b6b53392cf",
    "id": "c1582d2ed650a9742a0564c1e4df",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": 0.547878623008728,
    "voiceclap": 0.09651559591293335,
    "fitness": 0.8157707099519023,
    "legacy_fitness": 0.8157707099519023,
    "target_cdf": 0.9896249303026692,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6138916015625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7854466546363599,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6138916015625,
        "normalized": 0.9896249303026692,
        "contribution": 0.4948124651513346
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.046633720397949,
        "normalized": 0.23136007854264456,
        "contribution": 0.05784001963566114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.5132646560668945,
        "normalized": 0.9557276577709561,
        "contribution": 0.14335914866564342
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8448362350463867,
        "normalized": 0.8943502118372078,
        "contribution": 0.08943502118372078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8e9942a00b1bf5f483f0409831e689c44d5cb37858367272683fb59a3dc5cebc",
    "id": "c1ed09f02b46c82ee319d8425045",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.7793264389038086,
    "wer": 0.0,
    "identity": 0.5827453136444092,
    "voiceclap": 0.03836037963628769,
    "fitness": 0.7839180693990407,
    "legacy_fitness": 0.7839180693990407,
    "target_cdf": 0.9390533101491965,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8185012340545654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7060774253583981,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8185012340545654,
        "normalized": 0.9390533101491965,
        "contribution": 0.46952665507459823
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8159499168395996,
        "normalized": 0.15865523732432257,
        "contribution": 0.03966380933108064
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.61924409866333,
        "normalized": 0.8206824197073732,
        "contribution": 0.12310236295610598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.153956890106201,
        "normalized": 0.7378459799661322,
        "contribution": 0.07378459799661323
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "33bc809b3446464ce9382f8ceec822ed1812473ea7dd58b3199ba96a17306d7f",
    "id": "c218aba0298eacb9716070476c28",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 3.9801597595214844,
    "wer": 0.0,
    "identity": 0.6362789869308472,
    "voiceclap": -0.15948720276355743,
    "fitness": 0.797699443130204,
    "legacy_fitness": 0.797699443130204,
    "target_cdf": 0.9427315386702796,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.850965976715088,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7009902960706988,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.850965976715088,
        "normalized": 0.9427315386702796,
        "contribution": 0.4713657693351398
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6463295221328735,
        "normalized": 0.11598248129215816,
        "contribution": 0.02899562032303954
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.774919033050537,
        "normalized": 0.8542158295589685,
        "contribution": 0.12813237443384526
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1099517345428467,
        "normalized": 0.7249653197867423,
        "contribution": 0.07249653197867423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a9b5f67eb151957f6094fdede2bbf98265604d4431426c529ab499eede9ae825",
    "id": "c24a26d447f902cce9421d807847",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.08172385394573212,
    "fitness": 0.8142135456914237,
    "legacy_fitness": 0.8142135456914237,
    "target_cdf": 0.9850674461751243,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.467799186706543,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7090731567681856,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.467799186706543,
        "normalized": 0.9850674461751243,
        "contribution": 0.49253372308756216
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.3545823097229004,
        "normalized": 0.0628538085933158,
        "contribution": 0.01571345214832895
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8481216430664062,
        "normalized": 0.642531757010651,
        "contribution": 0.0642531757010651
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fda12f7c3898297f27a4c0a70e2511dbaaabce2d396a3accd4fe5e6535e4fb2b",
    "id": "c24b00acbc48482bb4b151bf2d20",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.9048471450805664,
    "wer": 0.0,
    "identity": 0.6909648776054382,
    "voiceclap": 0.036881223320961,
    "fitness": 0.8189464952412056,
    "legacy_fitness": 0.8189464952412056,
    "target_cdf": 0.9986501029854629,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.32811975479126,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8374828935600394,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.32811975479126,
        "normalized": 0.9986501029854629,
        "contribution": 0.49932505149273143
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.399444341659546,
        "normalized": 0.37140856111194015,
        "contribution": 0.09285214027798504
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.153757095336914,
        "normalized": 0.9882628216579012,
        "contribution": 0.1482394232486852
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.566519260406494,
        "normalized": 0.9706627854063763,
        "contribution": 0.09706627854063764
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "87a16ade01c9f9df57a5bedf6d4aed63b5ef36492cdea97299c6674b1e892fe8",
    "id": "c26609e2a6c2a6311a5da7cc257a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.038877829909324646,
    "fitness": 0.8249726935543273,
    "legacy_fitness": 0.8249726935543273,
    "target_cdf": 0.9542463867500731,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9645934104919434,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6778820958376642,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9645934104919434,
        "normalized": 0.9542463867500731,
        "contribution": 0.47712319337503656
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6463295221328735,
        "normalized": 0.11598248129215816,
        "contribution": 0.02899562032303954
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5614216327667236,
        "normalized": 0.8070803373590341,
        "contribution": 0.12106205060385511
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4564766883850098,
        "normalized": 0.5070123153573299,
        "contribution": 0.050701231535733
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "33282970cda503297db332f22649ee565660bda2885d983f4980b90a4e50f692",
    "id": "c26c45b84d8b906e54fc46b2cb45",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 3.8420867919921875,
    "wer": 0.0,
    "identity": 0.6444262266159058,
    "voiceclap": 0.06780778616666794,
    "fitness": 0.7961560816186836,
    "legacy_fitness": 0.7961560816186836,
    "target_cdf": 0.9495623609658421,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.915895938873291,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7129526870845684,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.915895938873291,
        "normalized": 0.9495623609658421,
        "contribution": 0.47478118048292106
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.079148292541504,
        "normalized": 0.7157671293545922,
        "contribution": 0.07157671293545922
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "76a56e8d2242b0f6f836b9a66d43151cae6086ddad92517e09153aa67e86c2b3",
    "id": "c2bfcff87b3dbea6d81d889b3959",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 3.823258876800537,
    "wer": 0.0,
    "identity": 0.7352830171585083,
    "voiceclap": 0.045630697160959244,
    "fitness": 0.7685462136779115,
    "legacy_fitness": 0.7685462136779115,
    "target_cdf": 0.9026628930794968,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.558781862258911,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7293876689427863,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.558781862258911,
        "normalized": 0.9026628930794968,
        "contribution": 0.4513314465397484
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.161975860595703,
        "normalized": 0.27373275044956896,
        "contribution": 0.06843318761239224
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.766023635864258,
        "normalized": 0.8524214149132157,
        "contribution": 0.12786321223698235
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4575917720794678,
        "normalized": 0.8175982255366333,
        "contribution": 0.08175982255366333
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "aace38e3de727177d528d5c405943a49e6208b6689b95a1c6d9f885aef0c9541",
    "id": "c31584cfaed109449c5e1dc5e8b5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": 0.5066837072372437,
    "voiceclap": 0.10739890486001968,
    "fitness": 0.7990650953987515,
    "legacy_fitness": 0.7990650953987515,
    "target_cdf": 0.9716911200959037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.191847801208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7373112062213696,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.191847801208496,
        "normalized": 0.9716911200959037,
        "contribution": 0.4858455600479519
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.184122562408447,
        "normalized": 0.9213282021786904,
        "contribution": 0.13819923032680356
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.395984649658203,
        "normalized": 0.8027715080715057,
        "contribution": 0.08027715080715057
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4d615dd2c3412b86ef286525754a5f7a34d358e605e9bb9b63cdad6bbf69f65d",
    "id": "c34b51b871a3e231677c58f1beb2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.086852550506592,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0590759739279747,
    "fitness": 0.7781759732199864,
    "legacy_fitness": 0.7781759732199864,
    "target_cdf": 0.8971763358976756,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5263171195983887,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6642098240547967,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5263171195983887,
        "normalized": 0.8971763358976756,
        "contribution": 0.4485881679488378
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.476912260055542,
        "normalized": 0.786101975965994,
        "contribution": 0.11791529639489909
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8239188194274902,
        "normalized": 0.634482821195086,
        "contribution": 0.0634482821195086
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "83696d5b2a5d21386f8ed9ac38f7ffd272536d73f6bc2410017178e4311b1a75",
    "id": "c3bd55c287a72cdb0ea4469da5a2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.015833841636776924,
    "fitness": 0.8281772960343221,
    "legacy_fitness": 0.8281772960343221,
    "target_cdf": 0.9954646088914592,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9223079681396484,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7780275406151345,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9223079681396484,
        "normalized": 0.9954646088914592,
        "contribution": 0.4977323044457296
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.886885643005371,
        "normalized": 0.97888519494114,
        "contribution": 0.146832779241171
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9064433574676514,
        "normalized": 0.9040003521739675,
        "contribution": 0.09040003521739676
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d6e681de987316ef7ee22a457430b418c9affca5b6c694cb1cb2a204044969d9",
    "id": "c3c310f25f0e30cf42ce808c4896",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 4.545003414154053,
    "wer": 0.0,
    "identity": 0.6435341835021973,
    "voiceclap": -0.1532132923603058,
    "fitness": 0.8265948856192484,
    "legacy_fitness": 0.8265948856192484,
    "target_cdf": 0.9542463867500731,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9645934104919434,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7436865347789139,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9645934104919434,
        "normalized": 0.9542463867500731,
        "contribution": 0.47712319337503656
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.19250750541687,
        "normalized": 0.2855588613245572,
        "contribution": 0.0713897153311393
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6414833068847656,
        "normalized": 0.8257492948479319,
        "contribution": 0.12386239422718978
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.070347309112549,
        "normalized": 0.7131123184554828,
        "contribution": 0.07131123184554829
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "067c830d52fffc60854bdc75bb7366a6b9023783569466a56b78ef9ba5966a42",
    "id": "c3d2bf1509c99726a2ad6b7934d5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.020888159051537514,
    "fitness": 0.8063459379026566,
    "legacy_fitness": 0.8063459379026566,
    "target_cdf": 0.9247120186784414,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.704874038696289,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7178510707641933,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.704874038696289,
        "normalized": 0.9247120186784414,
        "contribution": 0.4623560093392207
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.009317398071289,
        "normalized": 0.21847802587457404,
        "contribution": 0.05461950646864351
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.685961961746216,
        "normalized": 0.835607112993103,
        "contribution": 0.12534106694896543
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.215564012527466,
        "normalized": 0.7553448800736362,
        "contribution": 0.07553448800736362
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "885f9a2ff1ee335ad6b97a42be741aa8b1c0e4cc186cdc27178a5a31e753e27c",
    "id": "c3de50fab92275ba08a524ea5348",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": 0.7856932282447815,
    "voiceclap": 0.06852054595947266,
    "fitness": 0.7506280262523612,
    "legacy_fitness": 0.7506280262523612,
    "target_cdf": 0.8542158920076929,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3071789741516113,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7321763072699621,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3071789741516113,
        "normalized": 0.8542158920076929,
        "contribution": 0.42710794600384644
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.460507869720459,
        "normalized": 0.3982758645377146,
        "contribution": 0.09956896613442864
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5970046520233154,
        "normalized": 0.8155238285113872,
        "contribution": 0.12232857427670807
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5191988945007324,
        "normalized": 0.8317082085497887,
        "contribution": 0.08317082085497887
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "61b9e7046f4b5f1d1c9f92b3dbf093458cbd2a21abad76e8cc88b6b8a1f85d4a",
    "id": "c4a9c13e9c4ad835354e58b0650b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 3.835810661315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0075986385345458984,
    "fitness": 0.8148743618645529,
    "legacy_fitness": 0.8148743618645529,
    "target_cdf": 0.9954646088914592,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9223079681396484,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7796216327058983,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9223079681396484,
        "normalized": 0.9954646088914592,
        "contribution": 0.4977323044457296
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9007601737976074,
        "normalized": 0.18343713514349452,
        "contribution": 0.04585928378587363
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.67338752746582,
        "normalized": 0.9673901905993718,
        "contribution": 0.14510852858990578
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.941647529602051,
        "normalized": 0.9092151588438939,
        "contribution": 0.0909215158843894
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "206eae772395acac517141be62771081e53fe7e15c39c42c27c4fd67bb4eef49",
    "id": "c505d89015371ebf8aa4e06de40c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.284721851348877,
    "genuineness": 4.2249250411987305,
    "wer": 0.0,
    "identity": 0.7228798866271973,
    "voiceclap": -0.004264632239937782,
    "fitness": 0.8250805207707197,
    "legacy_fitness": 0.8250805207707197,
    "target_cdf": 0.9838478622074346,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4353346824645996,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.769550803061795,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4353346824645996,
        "normalized": 0.9838478622074346,
        "contribution": 0.4919239311037173
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.066988229751587,
        "normalized": 0.2385614708758641,
        "contribution": 0.059640367718966025
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8460850715637207,
        "normalized": 0.8680429242459118,
        "contribution": 0.13020643863688677
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7480251789093018,
        "normalized": 0.877800656022249,
        "contribution": 0.0877800656022249
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a08ad8ab6069fcf5327053f489dcaf2ae32205b12b17e687ec1417106a6cdbb2",
    "id": "c6313c8974daf298909f2176294d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.499174118041992,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.012259036302566528,
    "fitness": 0.7584664915052657,
    "legacy_fitness": 0.7584664915052657,
    "target_cdf": 0.8487882165208668,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.282830238342285,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7083992762125458,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.282830238342285,
        "normalized": 0.8487882165208668,
        "contribution": 0.4243941082604334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3282036781311035,
        "normalized": 0.3408456226113612,
        "contribution": 0.0852114056528403
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4813601970672607,
        "normalized": 0.7872381937406205,
        "contribution": 0.11808572906109308
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4135866165161133,
        "normalized": 0.807080332381791,
        "contribution": 0.0807080332381791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a293d5dd2459c93c1241d544033da45548c2709a20bd8e7fd407f1fb74fee9e1",
    "id": "c6394483f0f37aa0508748da6caa",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 3.71028995513916,
    "wer": 0.0,
    "identity": 0.7998882532119751,
    "voiceclap": 0.03277000039815903,
    "fitness": 0.7830272183782278,
    "legacy_fitness": 0.7830272183782278,
    "target_cdf": 0.9390533101491965,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8185012340545654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7049760252395696,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8185012340545654,
        "normalized": 0.9390533101491965,
        "contribution": 0.46952665507459823
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.810502052307129,
        "normalized": 0.8612466639446046,
        "contribution": 0.1291869995916907
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.136354923248291,
        "normalized": 0.7327310553381711,
        "contribution": 0.0732731055338171
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "38f9e26129582b55eb54519075d0fbc2e71610b127ab35469eac97da8acdc9c4",
    "id": "c64ada6ec27a8b82c19596cd26a7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.12477970123291016,
    "fitness": 0.8097132768813857,
    "legacy_fitness": 0.8097132768813857,
    "target_cdf": 0.9990117801396228,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.425514221191406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8095260613288477,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.425514221191406,
        "normalized": 0.9990117801396228,
        "contribution": 0.4995058900698114
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.134836435317993,
        "normalized": 0.26342804068679404,
        "contribution": 0.06585701017169851
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.296089172363281,
        "normalized": 0.99159014578754,
        "contribution": 0.148738521868131
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.337693214416504,
        "normalized": 0.9542463921920685,
        "contribution": 0.09542463921920685
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "773199f69d06d514d92155febc89e4b820d58172cda89361b3598678315fbd0c",
    "id": "c658efd6cbaeb8f6e7a103cdd281",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.432034492492676,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07382342219352722,
    "fitness": 0.834961334844543,
    "legacy_fitness": 0.834961334844543,
    "target_cdf": 0.9887331723771762,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5814266204833984,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7675926391506198,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5814266204833984,
        "normalized": 0.9887331723771762,
        "contribution": 0.4943665861885881
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0262794494628906,
        "normalized": 0.2242816522895154,
        "contribution": 0.05607041307237885
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.139644145965576,
        "normalized": 0.9154342857630313,
        "contribution": 0.1373151428644547
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.378382682800293,
        "normalized": 0.7984049702519815,
        "contribution": 0.07984049702519816
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ffcd724b10b69fd9abd4be47a16b2d206d11cba3ee1e982ef5172a29791b0891",
    "id": "c681279feb60cf0ae47a5d3969e9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.634199142456055,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1539613902568817,
    "fitness": 0.7767768905857946,
    "legacy_fitness": 0.7767768905857946,
    "target_cdf": 0.8629676751891268,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.347759962081909,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6531259858142229,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.347759962081909,
        "normalized": 0.8629676751891268,
        "contribution": 0.4314838375945634
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.161113977432251,
        "normalized": 0.6969418429694256,
        "contribution": 0.10454127644541385
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1627578735351562,
        "normalized": 0.740384500634085,
        "contribution": 0.0740384500634085
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b105debc42e2b3ffa849b76d9b0065a84b3f6c6049bb3f04be144cd7b16bf6a8",
    "id": "c6a9f78cf50460b7df1a607bbb58",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1473560929298401,
    "fitness": 0.7972386684454926,
    "legacy_fitness": 0.7972386684454926,
    "target_cdf": 0.9142167104159689,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6318278312683105,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6689839792924018,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6318278312683105,
        "normalized": 0.9142167104159689,
        "contribution": 0.45710835520798443
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4235379695892334,
        "normalized": 0.7721946200719969,
        "contribution": 0.11582919301079952
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.749110221862793,
        "normalized": 0.6092418912448665,
        "contribution": 0.060924189124486654
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3479d0064a62422aa9573802d6621df2f272935c4b0ce4df31edd4a1d9ecf462",
    "id": "c71f7859245595be6a2ed8d534a0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 4.011539936065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0019364096224308014,
    "fitness": 0.817776351412407,
    "legacy_fitness": 0.817776351412407,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7224588544995756,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5038483142852783,
        "normalized": 0.08701389328623654,
        "contribution": 0.021753473321559136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.184122562408447,
        "normalized": 0.9213282021786904,
        "contribution": 0.13819923032680356
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.079148292541504,
        "normalized": 0.7157671293545922,
        "contribution": 0.07157671293545922
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f4d04b0a1726da18848845d7bd106bd1171e6402dbd3888cd91321ffaac18471",
    "id": "c82279c25f64b6282abb6e63a98e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09405922144651413,
    "fitness": 0.8254494720408061,
    "legacy_fitness": 0.8254494720408061,
    "target_cdf": 0.9979798622347391,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.1982598304748535,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7899931953262849,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.1982598304748535,
        "normalized": 0.9979798622347391,
        "contribution": 0.49898993111736956
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9720008373260498,
        "normalized": 0.206019001985337,
        "contribution": 0.05150475049633425
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.993634223937988,
        "normalized": 0.9832067001034072,
        "contribution": 0.14748100501551106
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.020856857299805,
        "normalized": 0.9201750869706999,
        "contribution": 0.09201750869707
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f5eae47879fde86310d0bd2f93f156af981fa510dc4aed819e50fb99240937a7",
    "id": "c839ef6edf9ca4beb542ba0d065a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.25904521346092224,
    "fitness": 0.7466826325358982,
    "legacy_fitness": 0.7466826325358982,
    "target_cdf": 0.8070803415183294,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.112389326095581,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6191350688078598,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.112389326095581,
        "normalized": 0.8070803415183294,
        "contribution": 0.4035401707591647
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9278994798660278,
        "normalized": 0.19185148211659525,
        "contribution": 0.04796287052914881
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1344265937805176,
        "normalized": 0.6887065547816701,
        "contribution": 0.10330598321725051
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8503220081329346,
        "normalized": 0.6432604430229577,
        "contribution": 0.06432604430229577
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6242414e2837e3861344239e450984d8dfced488b999ca14768a501dd96f61a8",
    "id": "c85d47280417c5857d8d7cd68943",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.268836498260498,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.11824655532836914,
    "fitness": 0.8183298294411225,
    "legacy_fitness": 0.8183298294411225,
    "target_cdf": 0.999905909034833,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.091044902801514,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8112227394102236,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.091044902801514,
        "normalized": 0.999905909034833,
        "contribution": 0.4999529545174165
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.134836435317993,
        "normalized": 0.26342804068679404,
        "contribution": 0.06585701017169851
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.491794586181641,
        "normalized": 0.9948039170931102,
        "contribution": 0.14922058756396653
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.43450403213501,
        "normalized": 0.9619218715714208,
        "contribution": 0.09619218715714209
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9b3f29ae4c90f864d8633731e25e178d9a9bce1001a4c83d5aa6544793ea88c8",
    "id": "c8d5452d48f827d61c5dcf4e0a8e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.444586753845215,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": -0.053465478122234344,
    "fitness": 0.768964668589309,
    "legacy_fitness": 0.768964668589309,
    "target_cdf": 0.9079315965425183,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5912468433380127,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7310887142136826,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5912468433380127,
        "normalized": 0.9079315965425183,
        "contribution": 0.45396579827125916
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3214190006256104,
        "normalized": 0.3379849988867171,
        "contribution": 0.08449624972167928
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3568201065063477,
        "normalized": 0.7541161908392326,
        "contribution": 0.11311742862588489
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3651812076568604,
        "normalized": 0.7950923759485932,
        "contribution": 0.07950923759485932
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "59db588be95b1c4a8ac99fb8f74c0aa92d24566d408713efd5f4c390dacfdabe",
    "id": "c9370c8a8b599591fec67a8cf838",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.268857479095459,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04043056070804596,
    "fitness": 0.8323869700299755,
    "legacy_fitness": 0.8323869700299755,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.738620156141433,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.979520797729492,
        "normalized": 0.8914683718087433,
        "contribution": 0.1337202557713115
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.167158365249634,
        "normalized": 0.741648985261567,
        "contribution": 0.0741648985261567
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d6ee8893b6133037764a7a8bd30237d1b60ce3f84ef480f0195a3123d69846aa",
    "id": "c93d14a99d8ba6320f47d9de18eb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 3.553389072418213,
    "wer": 0.0,
    "identity": 0.5436865091323853,
    "voiceclap": 0.07932478934526443,
    "fitness": 0.7917390659346293,
    "legacy_fitness": 0.7917390659346293,
    "target_cdf": 0.9755090862302205,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.256777286529541,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8257793856520614,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.256777286529541,
        "normalized": 0.9755090862302205,
        "contribution": 0.48775454311511024
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.491039514541626,
        "normalized": 0.4119015661257916,
        "contribution": 0.1029753915314479
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.406516075134277,
        "normalized": 0.9462328279462877,
        "contribution": 0.14193492419194315
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.1088666915893555,
        "normalized": 0.9311452681356018,
        "contribution": 0.0931145268135602
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "15565469724665b1b2352c30f78959ca0e37664acc6eb7b8ff3d1e67ce79529d",
    "id": "c957809d230ecf5ed93290a71dc8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 4.6203155517578125,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.118742436170578,
    "fitness": 0.7169523885665476,
    "legacy_fitness": 0.7169523885665476,
    "target_cdf": 0.8394469613443116,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.242249011993408,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.639451476356159,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.242249011993408,
        "normalized": 0.8394469613443116,
        "contribution": 0.4197234806721558
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.10091233253479,
        "normalized": 0.2508316293237907,
        "contribution": 0.06270790733094768
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0766043663024902,
        "normalized": 0.670540497614825,
        "contribution": 0.10058107464222375
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.619295358657837,
        "normalized": 0.5643901371083175,
        "contribution": 0.05643901371083176
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b20c4dffd846a0b7fa09094833ddb196a5f8d0c7495be5339b5cee8d4c955842",
    "id": "c981586302f57ccd4507a869277c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.062326908111572,
    "genuineness": 3.5690791606903076,
    "wer": 0.0,
    "identity": 0.6916808485984802,
    "voiceclap": 0.06370501220226288,
    "fitness": 0.7984700768796618,
    "legacy_fitness": 0.7984700768796618,
    "target_cdf": 0.9964005449628859,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.003470420837402,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7862703845037892,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.003470420837402,
        "normalized": 0.9964005449628859,
        "contribution": 0.49820027248144294
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9143297672271729,
        "normalized": 0.18761514602105922,
        "contribution": 0.046903786505264805
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.064799785614014,
        "normalized": 0.9856469755282963,
        "contribution": 0.14784704632924445
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7ae61d63e468fe09a0beac07524e273c43947c98bcaefbbd84ce5c9220eb5d7c",
    "id": "c9896bccf1ecf660d4c45e6710bd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.6412534713745117,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.13960321247577667,
    "fitness": 0.8086039819763581,
    "legacy_fitness": 0.8086039819763581,
    "target_cdf": 0.9999312159312725,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.172207355499268,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.850848907838229,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.172207355499268,
        "normalized": 0.9999312159312725,
        "contribution": 0.49996560796563627
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.496976137161255,
        "normalized": 0.41456390432293383,
        "contribution": 0.10364097608073346
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.616334915161133,
        "normalized": 0.996228561583924,
        "contribution": 0.1494342842375886
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.707335472106934,
        "normalized": 0.9780803955427062,
        "contribution": 0.09780803955427063
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "67b833daf7b9b7d62d2e90461fab72d4630f451d6aba9e0e8426235b35bc6895",
    "id": "c99afb4d1c229514bfc9bdff7be1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.18726921081543,
    "wer": 0.0,
    "identity": 0.5868337750434875,
    "voiceclap": -0.02080519124865532,
    "fitness": 0.8275812309462424,
    "legacy_fitness": 0.8275812309462424,
    "target_cdf": 0.9915901345965457,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6950535774230957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7512409695822245,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6950535774230957,
        "normalized": 0.9915901345965457,
        "contribution": 0.49579506729827283
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.826127290725708,
        "normalized": 0.16150748713874752,
        "contribution": 0.04037687178468688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3255765438079834,
        "normalized": 0.7849622466803536,
        "contribution": 0.07849622466803537
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6974e339b2ec96e9ffc522c6a759f4bf0d0190a8386afb9f26d365f0833e84aa",
    "id": "ca08c213efef8ded44fa81248cd2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.19384190440177917,
    "fitness": 0.8093325362569509,
    "legacy_fitness": 0.8093325362569509,
    "target_cdf": 0.9390533101491965,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8185012340545654,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6845136252706666,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8185012340545654,
        "normalized": 0.9390533101491965,
        "contribution": 0.46952665507459823
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6056206226348877,
        "normalized": 0.10708370011583324,
        "contribution": 0.02677092502895831
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.525838851928711,
        "normalized": 0.7984049796580064,
        "contribution": 0.11976074694870095
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9779367446899414,
        "normalized": 0.6845529821840911,
        "contribution": 0.06845529821840911
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "be90c44a9e578fd3c2cf68fa28a2171f4a4a68c7fb5c7223174e3a7048bc9040",
    "id": "ca0d7d2be1b0f2e635f5f52da61d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.149696350097656,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": 0.5429689884185791,
    "voiceclap": -0.05423882603645325,
    "fitness": 0.8168119424792778,
    "legacy_fitness": 0.8168119424792778,
    "target_cdf": 0.9599408456112248,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0295231342315674,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6872407624729542,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0295231342315674,
        "normalized": 0.9599408456112248,
        "contribution": 0.4799704228056124
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5347344875335693,
        "normalized": 0.8005954285570323,
        "contribution": 0.12008931428355483
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.624795913696289,
        "normalized": 0.5663118737843122,
        "contribution": 0.05663118737843123
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "47cf9a19cffda7cb9c0c5f77b605d9b758ba6abc5c2d285508dc63624efed3fe",
    "id": "ca0e862d0d994d8bfb7c3e03fb67",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.09409761428833,
    "genuineness": 4.281409740447998,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": 0.07534829527139664,
    "fitness": 0.793224358648558,
    "legacy_fitness": 0.793224358648558,
    "target_cdf": 0.9796648197280255,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.337939739227295,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8378874832920367,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.337939739227295,
        "normalized": 0.9796648197280255,
        "contribution": 0.4898324098640128
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.574153423309326,
        "normalized": 0.449488735064188,
        "contribution": 0.112372183766047
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.3709330558776855,
        "normalized": 0.9427315543087618,
        "contribution": 0.14140973314631425
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.214478969573975,
        "normalized": 0.9427315651566268,
        "contribution": 0.09427315651566269
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8f645906a5b2edbc7d555e232d52e632051ef8d1888710a19785264b835f4392",
    "id": "ca1f5898ea689b73adaae77cbb2e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.545003414154053,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.03061683103442192,
    "fitness": 0.7883756137367388,
    "legacy_fitness": 0.7883756137367388,
    "target_cdf": 0.8885304563197098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4776196479797363,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6659113192687114,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4776196479797363,
        "normalized": 0.8885304563197098,
        "contribution": 0.4442652281598549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9855704307556152,
        "normalized": 0.2105000137670302,
        "contribution": 0.05262500344175755
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.125530958175659,
        "normalized": 0.6859401050066158,
        "contribution": 0.10289101575099237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9053282737731934,
        "normalized": 0.6613007191610654,
        "contribution": 0.06613007191610654
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2de720275ba300bf14004e55afd79e50d0313f1d14f0863495a2ae61e17edf32",
    "id": "ca3b7f54c2d4b3eda5ee7860b64c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.570657730102539,
    "genuineness": 4.595211505889893,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.01240556314587593,
    "fitness": 0.847005535584634,
    "legacy_fitness": 0.847005535584634,
    "target_cdf": 0.98254389250665,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.402869701385498,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6950354537871135,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.402869701385498,
        "normalized": 0.98254389250665,
        "contribution": 0.491271946253325
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.32744300365448,
        "normalized": 0.05908511838886088,
        "contribution": 0.01477127959721522
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.07737398147583,
        "normalized": 0.9066346475271524,
        "contribution": 0.13599519712907285
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.5213840007781982,
        "normalized": 0.5299703080750044,
        "contribution": 0.05299703080750045
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "146c4cd4c7d507aed787442ae3f9dcf3360eb2775226f61d7fe763357321d2dd",
    "id": "ca78d909ff5b0966545a6329d9d0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.055472373962402,
    "wer": 0.0,
    "identity": 0.3615086078643799,
    "voiceclap": -0.06956811994314194,
    "fitness": 0.8229944582441129,
    "legacy_fitness": 0.8229944582441129,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8057500598844098,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2094695568084717,
        "normalized": 0.29223234626397687,
        "contribution": 0.07305808656599422
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.459890365600586,
        "normalized": 0.951164403242087,
        "contribution": 0.14267466048631303
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.161672592163086,
        "normalized": 0.9371461719429255,
        "contribution": 0.09371461719429255
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f5a1aa8cf61739546a95418a65aeb39fce4a28aa43543bb8f262fa5627814029",
    "id": "ca8f5410065ec0739a4d2abc85f8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.111956596374512,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.012480376288294792,
    "fitness": 0.8222589692948854,
    "legacy_fitness": 0.8222589692948854,
    "target_cdf": 0.9915901345965457,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6950535774230957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7620941663570672,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6950535774230957,
        "normalized": 0.9915901345965457,
        "contribution": 0.49579506729827283
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.228601455688477,
        "normalized": 0.9269054845040974,
        "contribution": 0.1390358226756146
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8184332847595215,
        "normalized": 0.890006464053272,
        "contribution": 0.0890006464053272
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a057454c858e7aa0ddd821a3bf202340b3b6d9086d9adf8febc6f2710adac6c5",
    "id": "ca9978d11a686923e2ff128b5639",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": 0.6407393217086792,
    "voiceclap": 0.0020136479288339615,
    "fitness": 0.7824489172772866,
    "legacy_fitness": 0.7824489172772866,
    "target_cdf": 0.8885304563197098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4776196479797363,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6508571142930558,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4776196479797363,
        "normalized": 0.8885304563197098,
        "contribution": 0.4442652281598549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1833531856536865,
        "normalized": 0.7037290732409538,
        "contribution": 0.10555936098614306
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.911929130554199,
        "normalized": 0.663441906270323,
        "contribution": 0.0663441906270323
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9d6058d49adb6e0f046f865081131a44c8f27e5960b2a1887ff818addbbd9366",
    "id": "caf64b576ed7063a2b4f8b1b0d02",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.243753433227539,
    "wer": 0.030303030303030304,
    "identity": null,
    "voiceclap": -0.12425479292869568,
    "fitness": 0.8108641862808345,
    "legacy_fitness": 0.8108641862808345,
    "target_cdf": 0.9981710548076552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.230724811553955,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7524937241273818,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.230724811553955,
        "normalized": 0.9981710548076552,
        "contribution": 0.4990855274038276
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.3086628913879395,
        "normalized": 0.9361753524593768,
        "contribution": 0.1404263028689065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3343775272369385,
        "normalized": 0.787238162630965,
        "contribution": 0.0787238162630965
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "eb79fd23473c0c481b488f9a6c87382a619bc92a5d5e40600c39379fe652ca95",
    "id": "cb47f07ee9ec5474ea35526f75db",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0385594516992569,
    "fitness": 0.8283369148176758,
    "legacy_fitness": 0.8283369148176758,
    "target_cdf": 0.9896249303026692,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6138916015625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7224450925094233,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6138916015625,
        "normalized": 0.9896249303026692,
        "contribution": 0.4948124651513346
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.58526611328125,
        "normalized": 0.10282367927847097,
        "contribution": 0.025705919819617742
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.952833652496338,
        "normalized": 0.8870402984992682,
        "contribution": 0.13305604477489022
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.991138219833374,
        "normalized": 0.688706627635808,
        "contribution": 0.06887066276358081
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "197f439c15c88fc5c09413b4bf7dbf4b300f03725b0e8b3fa59828cb18e06985",
    "id": "cb85f16eac79a61abc900686682a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.18531018495559692,
    "fitness": 0.8172445853221395,
    "legacy_fitness": 0.8172445853221395,
    "target_cdf": 0.961270594530896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.045755386352539,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6934905499858022,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.045755386352539,
        "normalized": 0.961270594530896,
        "contribution": 0.480635297265448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6666839122772217,
        "normalized": 0.12062361006755057,
        "contribution": 0.030155902516887642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6592745780944824,
        "normalized": 0.8297365948158377,
        "contribution": 0.12446048922237565
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.671001434326172,
        "normalized": 0.5823886098109083,
        "contribution": 0.05823886098109084
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "046fc7a732ab383aad00ec007b5bf26b762dabe79a1c0d7fffdb522e6ae18b8e",
    "id": "cbc060ef00474bb62b8f99a8fe10",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.287685871124268,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1333332061767578,
    "fitness": 0.8190886459264647,
    "legacy_fitness": 0.8190886459264647,
    "target_cdf": 0.9650455080496254,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0944528579711914,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7554279363359475,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0944528579711914,
        "normalized": 0.9650455080496254,
        "contribution": 0.4825227540248127
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0568110942840576,
        "normalized": 0.23494557063169752,
        "contribution": 0.05873639265792438
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8016064167022705,
        "normalized": 0.8595110162463788,
        "contribution": 0.1289266524369568
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6160099506378174,
        "normalized": 0.852421372162536,
        "contribution": 0.08524213721625361
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "2bdc77ce1eeb959e0d54dfbb7934a04ae22ca2baa9c47897f2a7f788037243f4",
    "id": "cbdedb7385d8dbf4aebea5e97de0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.074300289154053,
    "wer": 0.0,
    "identity": 0.603151798248291,
    "voiceclap": 0.09447704255580902,
    "fitness": 0.7804282971525778,
    "legacy_fitness": 0.7804282971525778,
    "target_cdf": 0.8999470925936282,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5425496101379395,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7008187756344689,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5425496101379395,
        "normalized": 0.8999470925936282,
        "contribution": 0.4499735462968141
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1144819259643555,
        "normalized": 0.25583161582232095,
        "contribution": 0.06395790395558024
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4813601970672607,
        "normalized": 0.7872381937406205,
        "contribution": 0.11808572906109308
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9889378547668457,
        "normalized": 0.6880159632098157,
        "contribution": 0.06880159632098158
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5c47022f9e13e51befb1d310020c8268da796c2940c81febc4d2817a228935a5",
    "id": "cc1abd88b90345cc79542ff0024f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.046441555023193,
    "genuineness": 4.356721878051758,
    "wer": 0.03333333333333333,
    "identity": 0.18427124619483948,
    "voiceclap": 0.019222650676965714,
    "fitness": 0.7403448738595316,
    "legacy_fitness": 0.7403448738595316,
    "target_cdf": 0.8663657806263168,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.36399245262146,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7002918218572916,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.36399245262146,
        "normalized": 0.8663657806263168,
        "contribution": 0.4331828903131584
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3672165870666504,
        "normalized": 0.35746827293550454,
        "contribution": 0.08936706823387613
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1966967582702637,
        "normalized": 0.7077676691653187,
        "contribution": 0.1061651503747978
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.079148292541504,
        "normalized": 0.7157671293545922,
        "contribution": 0.07157671293545922
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f18678f2420849e2106196d4b6c456d735a77e37caf468007e404222218bddd6",
    "id": "cc5b98ff99358d320a4bbeb56f91",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": 0.4577191472053528,
    "voiceclap": -0.11049407720565796,
    "fitness": 0.834513471228712,
    "legacy_fitness": 0.834513471228712,
    "target_cdf": 0.988262823627061,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5651941299438477,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7347897034732848,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5651941299438477,
        "normalized": 0.988262823627061,
        "contribution": 0.4941314118135305
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7209625244140625,
        "normalized": 0.13363422338747727,
        "contribution": 0.03340855584686932
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.979520797729492,
        "normalized": 0.8914683718087433,
        "contribution": 0.1337202557713115
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.145155906677246,
        "normalized": 0.7352948004157347,
        "contribution": 0.07352948004157348
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "84dd2d561b4ce214accf768d958bf32cc991411ba53907395c340ece890da58a",
    "id": "cce9de38a062c91434ad02eb4f2d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.6851859092712402,
    "wer": 0.0,
    "identity": 0.7398280501365662,
    "voiceclap": 0.09759558737277985,
    "fitness": 0.8043222083866055,
    "legacy_fitness": 0.8043222083866055,
    "target_cdf": 0.98254389250665,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.402869701385498,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7309919135111961,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.402869701385498,
        "normalized": 0.98254389250665,
        "contribution": 0.491271946253325
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.259568929672241,
        "normalized": 0.7674481251957699,
        "contribution": 0.07674481251957699
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1dc1716b258df1f5c5e156c967b0e40d5d2f87af186557ded47de5e2ae3adaae",
    "id": "cd53d77129f9c00fdf3afadacfad",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 4.017816066741943,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.06386171281337738,
    "fitness": 0.8206624501040387,
    "legacy_fitness": 0.8206624501040387,
    "target_cdf": 0.9811510787587977,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3704047203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7234875271280466,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3704047203063965,
        "normalized": 0.9811510787587977,
        "contribution": 0.49057553937939885
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5716965198516846,
        "normalized": 0.1000529371132764,
        "contribution": 0.0250132342783191
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.086269855499268,
        "normalized": 0.9079316126330063,
        "contribution": 0.13618974189495092
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0835487842559814,
        "normalized": 0.7170901157537775,
        "contribution": 0.07170901157537775
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "42dd394703b9b8f8745649e0e9703d963639f89f19ceaed232f6441427ce4d7c",
    "id": "cd8c2f9fdfd69802905643672926",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": 0.5336751341819763,
    "voiceclap": 0.055153973400592804,
    "fitness": 0.8459105766134407,
    "legacy_fitness": 0.8459105766134407,
    "target_cdf": 0.9985049319896988,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.295654773712158,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7353068952922163,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.295654773712158,
        "normalized": 0.9985049319896988,
        "contribution": 0.4992524659948494
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5038483142852783,
        "normalized": 0.08701389328623654,
        "contribution": 0.021753473321559136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.406516075134277,
        "normalized": 0.9462328279462877,
        "contribution": 0.14193492419194315
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.105551242828369,
        "normalized": 0.7236603178386457,
        "contribution": 0.07236603178386457
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ee914e5f9c70fbc11a7c146e049dbcbccf321d4d09b4206a6af9a2c26d82ab99",
    "id": "ce00e70472f801d6761115d356fe",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.2249250411987305,
    "wer": 0.07142857142857142,
    "identity": null,
    "voiceclap": -0.15545830130577087,
    "fitness": 0.7509590056139577,
    "legacy_fitness": 0.7509590056139577,
    "target_cdf": 0.9527255219125881,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9483609199523926,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6696126837051565,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9483609199523926,
        "normalized": 0.9527255219125881,
        "contribution": 0.47636276095629404
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.598835825920105,
        "normalized": 0.10564978306864803,
        "contribution": 0.026412445767162007
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2144882678985596,
        "normalized": 0.7131122892698544,
        "contribution": 0.10696684339047816
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.71830677986145,
        "normalized": 0.5987063359122227,
        "contribution": 0.05987063359122227
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b738b6fe363f8c3a94ba25c21a4c28c33a60f864c5e5991c89ac702152924f83",
    "id": "ce2b37c164663621646bc2ddc311",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 3.986435890197754,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.014612291008234024,
    "fitness": 0.8152839329660394,
    "legacy_fitness": 0.8152839329660394,
    "target_cdf": 0.9745975945036527,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2405447959899902,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7123643297198763,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2405447959899902,
        "normalized": 0.9745975945036527,
        "contribution": 0.48729879725182634
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5784813165664673,
        "normalized": 0.10143141696818753,
        "contribution": 0.02535785424204688
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8994593620300293,
        "normalized": 0.8778006492473432,
        "contribution": 0.13167009738710148
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9647350311279297,
        "normalized": 0.6803758083890165,
        "contribution": 0.06803758083890164
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5c0820601a3326bf8b26973d45f452880067d0adef5a6b50a3de35f79ab82c26",
    "id": "cea8225b84f6eb2dfe153db19a5b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 3.6977379322052,
    "wer": 0.0,
    "identity": 0.6795995831489563,
    "voiceclap": 0.09669888019561768,
    "fitness": 0.7988225982739836,
    "legacy_fitness": 0.7988225982739836,
    "target_cdf": 0.9685130012341955,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1431503295898438,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7756052000408389,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1431503295898438,
        "normalized": 0.9685130012341955,
        "contribution": 0.48425650061709774
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2026846408843994,
        "normalized": 0.28955421544612325,
        "contribution": 0.07238855386153081
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5368008613586426,
        "normalized": 0.8356070969154066,
        "contribution": 0.08356070969154067
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "822408d859d0f333a379e387ab4cb0dcda994817cfc7e580cb6a227a1328fea8",
    "id": "cee63657a1af91e469142d929d96",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.15560518205165863,
    "fitness": 0.8062114252975141,
    "legacy_fitness": 0.8062114252975141,
    "target_cdf": 0.9557276754241297,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.980825901031494,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6933676596604943,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.980825901031494,
        "normalized": 0.9557276754241297,
        "contribution": 0.47786383771206487
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.427985668182373,
        "normalized": 0.7733726160960801,
        "contribution": 0.116005892414412
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.942732572555542,
        "normalized": 0.67336307136718,
        "contribution": 0.067336307136718
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b09bc9311a7f40d6068292742d30f4cc77ca4a7d99e9ab5c813dd9277426c610",
    "id": "cef5b3877eb34e4fba11240eec96",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": 0.5866222381591797,
    "voiceclap": 0.05079934000968933,
    "fitness": 0.7685486910987934,
    "legacy_fitness": 0.7685486910987934,
    "target_cdf": 0.9092151558893999,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.599363088607788,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6785435227079999,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.599363088607788,
        "normalized": 0.9092151558893999,
        "contribution": 0.45460757794469997
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7277473211288452,
        "normalized": 0.1353259560349948,
        "contribution": 0.0338314890087487
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5836610794067383,
        "normalized": 0.8123848387569603,
        "contribution": 0.12185772581354404
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9713358879089355,
        "normalized": 0.6824672994100721,
        "contribution": 0.06824672994100721
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d90546fc6583f4a847eadc920bbe0b302fdbe5f2ea4bee026a5fce065bcaae86",
    "id": "cf21c6850b79e29c7b97637ac584",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.027126196771860123,
    "fitness": 0.8179609611697054,
    "legacy_fitness": 0.8179609611697054,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7637036050453336,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7888107299804688,
        "normalized": 0.1512118069914083,
        "contribution": 0.03780295174785207
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.566638946533203,
        "normalized": 0.9599408341035309,
        "contribution": 0.14399112511552964
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.668816089630127,
        "normalized": 0.8629676828711841,
        "contribution": 0.08629676828711841
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d96472c3424accbad1cab460eecb152ca0f7530dc4818340549c697688e30001",
    "id": "cf5143074dae9792be3c2872b5c9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.44,
    "cfg": 2,
    "generation": null,
    "ce": 6.459460258483887,
    "genuineness": 4.20609712600708,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11763749271631241,
    "fitness": 0.8028280639537156,
    "legacy_fitness": 0.8028280639537156,
    "target_cdf": 0.9321750211493136,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.761687755584717,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7016571376065672,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.761687755584717,
        "normalized": 0.9321750211493136,
        "contribution": 0.4660875105746568
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.650379180908203,
        "normalized": 0.8277503599224988,
        "contribution": 0.12416255398837481
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1319541931152344,
        "normalized": 0.7314444306568307,
        "contribution": 0.07314444306568307
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "44c632d6a4c7c3df6ac9d90b6d852af3b727418677a25dfc9d1be2c8110af901",
    "id": "cf5ce69f6d6a94372e26640456b3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": 0.6599034070968628,
    "voiceclap": -0.018525658175349236,
    "fitness": 0.7833970965022261,
    "legacy_fitness": 0.7833970965022261,
    "target_cdf": 0.9104853512991299,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6074790954589844,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6793290911762705,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6074790954589844,
        "normalized": 0.9104853512991299,
        "contribution": 0.45524267564956494
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.694857597351074,
        "normalized": 0.8375344060053678,
        "contribution": 0.12563016090080517
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.997738838195801,
        "normalized": 0.6907744818671719,
        "contribution": 0.0690774481867172
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5a1fac853aa3e65543df5103e6632be0936d5114e9ff42ba2c52e5f6860e77a0",
    "id": "d024c5dac7a7408e4648ba3ee9fa",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.13078498840332,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": 0.025152437388896942,
    "fitness": 0.7983235399369322,
    "legacy_fitness": 0.7983235399369322,
    "target_cdf": 0.9989025172806805,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.393049240112305,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7503881209107596,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.393049240112305,
        "normalized": 0.9989025172806805,
        "contribution": 0.49945125864034023
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4970635175704956,
        "normalized": 0.08578327395609947,
        "contribution": 0.021445818489024868
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.48657751083374,
        "normalized": 0.9534909337644679,
        "contribution": 0.1430236400646702
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.677617073059082,
        "normalized": 0.8646740371672423,
        "contribution": 0.08646740371672423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8d3745951face1c2a63f0c620d1db2683d239f643993e82a8b433c8486bee5a4",
    "id": "d0782ab73ddd3a709d8c9c0694ec",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07428858429193497,
    "fitness": 0.832427879910752,
    "legacy_fitness": 0.832427879910752,
    "target_cdf": 0.9891870513381258,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.597659111022949,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7163111603333052,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.597659111022949,
        "normalized": 0.9891870513381258,
        "contribution": 0.4945935256690629
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6259750127792358,
        "normalized": 0.11146953771947371,
        "contribution": 0.027867384429868428
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.766023635864258,
        "normalized": 0.8524214149132157,
        "contribution": 0.12786321223698235
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.900927782058716,
        "normalized": 0.6598703799739153,
        "contribution": 0.06598703799739153
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "04794690cf9208b86eca5c1d55fd5bd5b4355f1050d20c012e48e5075dcfea12",
    "id": "d0828501eeb294d2957c30c5d144",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.051873281598091125,
    "fitness": 0.7412342744858142,
    "legacy_fitness": 0.7412342744858142,
    "target_cdf": 0.8113312001607969,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.128621816635132,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6318331386010468,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.128621816635132,
        "normalized": 0.8113312001607969,
        "contribution": 0.4056656000803984
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.043241500854492,
        "normalized": 0.23017181441662055,
        "contribution": 0.057542953604155136
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.036573648452759,
        "normalized": 0.6577205162795667,
        "contribution": 0.09865807744193501
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0263421535491943,
        "normalized": 0.6996650747455826,
        "contribution": 0.06996650747455826
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "af059616076e34fd3953bd6082c7d8598d5e00610e06d386bb0d2e4318eb8cbd",
    "id": "d09225346d898fd787e9713d2c5d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 3.9676074981689453,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.027273761108517647,
    "fitness": 0.8203275400116833,
    "legacy_fitness": 0.8203275400116833,
    "target_cdf": 0.9977706984112247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.165794849395752,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8578479815460691,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.165794849395752,
        "normalized": 0.9977706984112247,
        "contribution": 0.49888534920561234
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5622799396514893,
        "normalized": 0.44408306097259703,
        "contribution": 0.11102076524314926
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.420629024505615,
        "normalized": 0.9937903346742238,
        "contribution": 0.14906855020113358
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 5.006569862365723,
        "normalized": 0.9887331689617396,
        "contribution": 0.09887331689617396
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "25598f38e4fc734d62facfa109930716628dc87a1e0a2bb49692aeb314c69292",
    "id": "d0b86a23338ea2f764189332749f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 3.986435890197754,
    "wer": 0.0,
    "identity": 0.43032306432724,
    "voiceclap": 0.039368607103824615,
    "fitness": 0.825283572816527,
    "legacy_fitness": 0.825283572816527,
    "target_cdf": 0.9981710548076552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.230724811553955,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7626793324402976,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.230724811553955,
        "normalized": 0.9981710548076552,
        "contribution": 0.4990855274038276
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.406516075134277,
        "normalized": 0.9462328279462877,
        "contribution": 0.14193492419194315
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7040200233459473,
        "normalized": 0.8697054632450133,
        "contribution": 0.08697054632450134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d208764cb7a1700f635749bbb19d6b325d199e94ba80f53c03078cc28b709807",
    "id": "d0b968b3a35e8fa71a261b92b31e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.143336772918701,
    "wer": 0.0,
    "identity": 0.6702951788902283,
    "voiceclap": -0.021481063216924667,
    "fitness": 0.7820949273475297,
    "legacy_fitness": 0.7820949273475297,
    "target_cdf": 0.8943502420521354,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.510084629058838,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.658214060096017,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.510084629058838,
        "normalized": 0.8943502420521354,
        "contribution": 0.4471751210260677
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.598835825920105,
        "normalized": 0.10564978306864803,
        "contribution": 0.026412445767162007
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.490255832672119,
        "normalized": 0.7894999685105402,
        "contribution": 0.11842499527658101
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9075284004211426,
        "normalized": 0.6620149802620625,
        "contribution": 0.06620149802620626
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "da3c7f7e4b15d30ac66817f5bb87cc2bf7b8196a1f77aebc9d3d880278057ca0",
    "id": "d0bcc0d028eeb333adc6ea98a08a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.570107460021973,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11581705510616302,
    "fitness": 0.8200421111126031,
    "legacy_fitness": 0.8200421111126031,
    "target_cdf": 0.9418287707049221,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8428497314453125,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7235201409805735,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8428497314453125,
        "normalized": 0.9418287707049221,
        "contribution": 0.47091438535246105
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1212668418884277,
        "normalized": 0.25835102183826425,
        "contribution": 0.06458775545956606
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4190900325775146,
        "normalized": 0.7710131225645452,
        "contribution": 0.11565196838468178
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.105551242828369,
        "normalized": 0.7236603178386457,
        "contribution": 0.07236603178386457
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1708551e6af3f7868132717f57e338c7ce98d7fd1c116b813b0a25e93ac2c23e",
    "id": "d11f4682255405cac1e8892e5473",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.229123115539551,
    "genuineness": 3.5722172260284424,
    "wer": 0.0,
    "identity": 0.6671259999275208,
    "voiceclap": 0.07254070043563843,
    "fitness": 0.7838575700932803,
    "legacy_fitness": 0.7838575700932803,
    "target_cdf": 0.9585742155738795,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0132906436920166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.779164754241893,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0132906436920166,
        "normalized": 0.9585742155738795,
        "contribution": 0.4792871077869397
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.189115047454834,
        "normalized": 0.2842329008639445,
        "contribution": 0.07105822521598612
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9328465461730957,
        "normalized": 0.9079316065734483,
        "contribution": 0.09079316065734483
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7a4181a68a143b0c5f5a2ceb5fbd1f41c22424b7dbe08156b34bcbe53008a54e",
    "id": "d1238d84af7f62034182dc5922c3",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.84,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.173524379730225,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.1166638657450676,
    "fitness": 0.8222008885919742,
    "legacy_fitness": 0.8222008885919742,
    "target_cdf": 0.9967235449909552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.035935401916504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7950635646371353,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.035935401916504,
        "normalized": 0.9967235449909552,
        "contribution": 0.4983617724954776
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.155190944671631,
        "normalized": 0.2711379669791497,
        "contribution": 0.06778449174478743
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.353141784667969,
        "normalized": 0.9409148986309699,
        "contribution": 0.1411372347946455
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7480251789093018,
        "normalized": 0.877800656022249,
        "contribution": 0.0877800656022249
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a9fe46b5ff0a0e4c4990ced66e4a262c0d5f8cfae8153c3e59520c6f1ff9e80d",
    "id": "d126268fdf50056f6eead78e8ac0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.046441555023193,
    "genuineness": 3.5439748764038086,
    "wer": 0.0,
    "identity": 0.7113246917724609,
    "voiceclap": 0.08400051295757294,
    "fitness": 0.798079124720588,
    "legacy_fitness": 0.798079124720588,
    "target_cdf": 0.9985049319896988,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.295654773712158,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8194114799893667,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.295654773712158,
        "normalized": 0.9985049319896988,
        "contribution": 0.4992524659948494
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.226431369781494,
        "normalized": 0.29897739616091534,
        "contribution": 0.07474434904022884
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.31387996673584,
        "normalized": 0.9919413784114786,
        "contribution": 0.14879120676172178
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.496110916137695,
        "normalized": 0.9662345819256669,
        "contribution": 0.0966234581925667
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "280ee97e3559484875ea4e550ba40d62b9aa86228a5eedd9fcf96a0760790fdb",
    "id": "d15b9476af5d33ee48f886915fb0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.582659721374512,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12441511452198029,
    "fitness": 0.8421485314718171,
    "legacy_fitness": 0.8421485314718171,
    "target_cdf": 0.9877755264344382,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.548961639404297,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.724181740090505,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.548961639404297,
        "normalized": 0.9877755264344382,
        "contribution": 0.4938877632172191
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7277473211288452,
        "normalized": 0.1353259560349948,
        "contribution": 0.0338314890087487
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7571277618408203,
        "normalized": 0.8506121673397089,
        "contribution": 0.12759182510095632
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.991138219833374,
        "normalized": 0.688706627635808,
        "contribution": 0.06887066276358081
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8658fba48e8a25f3c95c84bfc1751ae7ef07ca7d0c3be2ac3d682cdfe13d95f7",
    "id": "d175ead2468f02a406e083d6eeae",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.095972441136837,
    "fitness": 0.7812811516392111,
    "legacy_fitness": 0.7812811516392111,
    "target_cdf": 0.8793763603106894,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.428922176361084,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6110877458049402,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.428922176361084,
        "normalized": 0.8793763603106894,
        "contribution": 0.4396881801553447
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6191902160644531,
        "normalized": 0.1099935309435226,
        "contribution": 0.02749838273588065
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1233069896698,
        "normalized": 0.6852468300972876,
        "contribution": 0.10278702451459314
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.183645248413086,
        "normalized": 0.41114158399121653,
        "contribution": 0.04111415839912166
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e23403b3b55123831c1ab2155d34703b3c7d621e6d8e8f4a9559118f07bdd2f9",
    "id": "d21812c0c4d6be7c5e377f815db9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.607763767242432,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1353769600391388,
    "fitness": 0.8126402231686819,
    "legacy_fitness": 0.8126402231686819,
    "target_cdf": 0.929049708969098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7373390197753906,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6501457363875833,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7373390197753906,
        "normalized": 0.929049708969098,
        "contribution": 0.464524854484549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8363044261932373,
        "normalized": 0.16439289419288416,
        "contribution": 0.04109822354822104
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0299019813537598,
        "normalized": 0.6555656036609419,
        "contribution": 0.09833484054914128
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.328861951828003,
        "normalized": 0.46187817805672027,
        "contribution": 0.04618781780567203
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "91cc44d3daf551809324eb78f1062a0eb7484502bef03754a9dbb368e36d9eec",
    "id": "d29bc1a4265716fe24dbcb3760ce",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.337893962860107,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.03678327798843384,
    "fitness": 0.8247357912600557,
    "legacy_fitness": 0.8247357912600557,
    "target_cdf": 0.9673902089874984,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.126917839050293,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7046455872985233,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.126917839050293,
        "normalized": 0.9673902089874984,
        "contribution": 0.4836951044937492
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7393362522125244,
        "normalized": 0.8469494732316728,
        "contribution": 0.1270424209847509
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.773313045501709,
        "normalized": 0.6174643942272378,
        "contribution": 0.06174643942272378
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d9f39a7cd0126b304ebff2727e4e8de0f16dd59ed42814d5832eccdf6cad50a4",
    "id": "d2b183d23731135b67c3563cdca0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.218649387359619,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04110664501786232,
    "fitness": 0.8174325659088941,
    "legacy_fitness": 0.8174325659088941,
    "target_cdf": 0.9599408456112248,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0295231342315674,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7066247535297433,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0295231342315674,
        "normalized": 0.9599408456112248,
        "contribution": 0.4799704228056124
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.768456220626831,
        "normalized": 0.1457841285203917,
        "contribution": 0.03644603213009792
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5614216327667236,
        "normalized": 0.8070803373590341,
        "contribution": 0.12106205060385511
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.999939203262329,
        "normalized": 0.6914624799017783,
        "contribution": 0.06914624799017784
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0a66db376e9517d3c099c11776ebf8d525e7122ced36532a2acfe25afb5e0381",
    "id": "d2c2d060d80ca29bedfdfa631558",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 3.6977379322052,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.08470868319272995,
    "fitness": 0.789848565492089,
    "legacy_fitness": 0.789848565492089,
    "target_cdf": 0.9557276754241297,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.980825901031494,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7599146559966785,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.980825901031494,
        "normalized": 0.9557276754241297,
        "contribution": 0.47786383771206487
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1416213512420654,
        "normalized": 0.26598556001729745,
        "contribution": 0.06649639000432436
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.837189197540283,
        "normalized": 0.8663657575315874,
        "contribution": 0.1299548636297381
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6336121559143066,
        "normalized": 0.8559956465055119,
        "contribution": 0.08559956465055119
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a7cc8a0b48db1fbcefa0f362fc5cae932d7e07e061f95d4f00edc08d15f00372",
    "id": "d2ec3790b6d79ee31e10f4705e70",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.469690799713135,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.08258572220802307,
    "fitness": 0.8331948854251755,
    "legacy_fitness": 0.8331948854251755,
    "target_cdf": 0.9788851773190603,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.321707248687744,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6839537841493609,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.321707248687744,
        "normalized": 0.9788851773190603,
        "contribution": 0.48944258865953016
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6531143188476562,
        "normalized": 0.11751522575673273,
        "contribution": 0.029378806439183183
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3879549503326416,
        "normalized": 0.7626471853600028,
        "contribution": 0.11439707780400041
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4574391841888428,
        "normalized": 0.5073531124664714,
        "contribution": 0.050735311246647145
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cb14739e32ad61df805429ef1fbb2da7dd607bfdb3c35262116cf27e2b8b4544",
    "id": "d2f2d153b570f6d19352093e6c28",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.260893821716309,
    "genuineness": 4.507347106933594,
    "wer": 0.0,
    "identity": 0.4874677062034607,
    "voiceclap": -0.16110759973526,
    "fitness": 0.8191096554813952,
    "legacy_fitness": 0.8191096554813952,
    "target_cdf": 0.9495623609658421,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.915895938873291,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7136471316658624,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.915895938873291,
        "normalized": 0.9495623609658421,
        "contribution": 0.47478118048292106
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.046633720397949,
        "normalized": 0.23136007854264456,
        "contribution": 0.05784001963566114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3879549503326416,
        "normalized": 0.7626471853600028,
        "contribution": 0.11439707780400041
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9207301139831543,
        "normalized": 0.6662885374327974,
        "contribution": 0.06662885374327975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ef55effd113efcda32532e07e19ff453b8629d437b199c814fddebb16206d035",
    "id": "d33433dcbae47ae35b96ff88c4e8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.181467056274414,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.039689987897872925,
    "fitness": 0.8247249139987738,
    "legacy_fitness": 0.8247249139987738,
    "target_cdf": 0.9992824581550183,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.522909164428711,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8095217748057782,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.522909164428711,
        "normalized": 0.9992824581550183,
        "contribution": 0.49964122907750913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1382288932800293,
        "normalized": 0.2647052295903163,
        "contribution": 0.06617630739757907
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.011425495147705,
        "normalized": 0.9838478542017792,
        "contribution": 0.14757717813026688
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.425703048706055,
        "normalized": 0.9612706020042315,
        "contribution": 0.09612706020042316
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "30c6c6d0606198b6906b229e60ff507ec397e6aac93755bacacd322bc02338a0",
    "id": "d387616b882e6dc52eb9c631526c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2,
    "generation": null,
    "ce": 6.229123115539551,
    "genuineness": 4.256305694580078,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": 0.03580084070563316,
    "fitness": 0.8010328755868666,
    "legacy_fitness": 0.8010328755868666,
    "target_cdf": 0.9952532227437393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9060754776000977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7946685079411306,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9060754776000977,
        "normalized": 0.9952532227437393,
        "contribution": 0.49762661137186964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1823301315307617,
        "normalized": 0.2815898456276642,
        "contribution": 0.07039746140691605
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.210809707641602,
        "normalized": 0.9247119875768577,
        "contribution": 0.13870679813652864
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.756826162338257,
        "normalized": 0.8793763702581625,
        "contribution": 0.08793763702581625
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c2e6cc2e2128b44baaffecc1549335b2d0f51dbe5c6ff07e48742fc7d0c5a998",
    "id": "d3c3ef6ac87e5652c070d52c95bf",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.078212261199951,
    "genuineness": 4.519899368286133,
    "wer": 0.03333333333333333,
    "identity": null,
    "voiceclap": -0.2115509808063507,
    "fitness": 0.7409453119288728,
    "legacy_fitness": 0.7409453119288728,
    "target_cdf": 0.8524214067919869,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.299062728881836,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5770489794277369,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.299062728881836,
        "normalized": 0.8524214067919869,
        "contribution": 0.42621070339599343
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6191902160644531,
        "normalized": 0.1099935309435226,
        "contribution": 0.02749838273588065
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.809732675552368,
        "normalized": 0.5820072855367588,
        "contribution": 0.08730109283051382
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.0340280532836914,
        "normalized": 0.3603880046534898,
        "contribution": 0.036038800465348984
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "abe4f18f674e7e0767739cfa25a8eda15e55797b7dbc0fc6c26e80f2ba974e0c",
    "id": "d3ccfddac2d255ef10119cfafdcf",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.1894097328186035,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": 0.7556030750274658,
    "voiceclap": -0.13806964457035065,
    "fitness": 0.8140810570630919,
    "legacy_fitness": 0.8140810570630919,
    "target_cdf": 0.9650455080496254,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0944528579711914,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7087661350925571,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0944528579711914,
        "normalized": 0.9650455080496254,
        "contribution": 0.4825227540248127
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7888107299804688,
        "normalized": 0.1512118069914083,
        "contribution": 0.03780295174785207
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4101943969726562,
        "normalized": 0.7686399083613955,
        "contribution": 0.11529598625420932
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1319541931152344,
        "normalized": 0.7314444306568307,
        "contribution": 0.07314444306568307
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6a1160b6a171a964a9821c5ab2b0c9c5d43117a5a395fe00263ffc9ffdada6b6",
    "id": "d435c0002fc83537df3b767d1f21",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.7081804275512695,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.27858540415763855,
    "fitness": 0.7151414930848322,
    "legacy_fitness": 0.7151414930848322,
    "target_cdf": 0.7197271863675201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.8161470890045166,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5977831699449832,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.8161470890045166,
        "normalized": 0.7197271863675201,
        "contribution": 0.35986359318376004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2942795753479004,
        "normalized": 0.3266368847639896,
        "contribution": 0.0816592211909974
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.9342727661132812,
        "normalized": 0.6241528877228036,
        "contribution": 0.09362293315842055
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.799715995788574,
        "normalized": 0.6263742241180521,
        "contribution": 0.06263742241180521
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "36436af1fee484170a21f6842a369de63db2d58956077557622285f76a80a45d",
    "id": "d439672137f8f83767540ec919b0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.582659721374512,
    "wer": 0.07142857142857142,
    "identity": 0.5844658613204956,
    "voiceclap": -0.10228510946035385,
    "fitness": 0.7450040393923535,
    "legacy_fitness": 0.7450040393923535,
    "target_cdf": 0.9104853512991299,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6074790954589844,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.643201140304044,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6074790954589844,
        "normalized": 0.9104853512991299,
        "contribution": 0.45524267564956494
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.98319935798645,
        "normalized": 0.6403427152594581,
        "contribution": 0.09605140728891871
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6291964054107666,
        "normalized": 0.5678481541642914,
        "contribution": 0.05678481541642914
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ae590ffaa488c712c7937531daecca39b66f82d31656f6a354680c1e6cc6fc66",
    "id": "d4c04802f91ea42553ecabd7e9c8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.898571252822876,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.08071736991405487,
    "fitness": 0.8230729161988763,
    "legacy_fitness": 0.8230729161988763,
    "target_cdf": 0.9994832593780891,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.620303630828857,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7962006411346426,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.620303630828857,
        "normalized": 0.9994832593780891,
        "contribution": 0.49974162968904456
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0228869915008545,
        "normalized": 0.22311395750311608,
        "contribution": 0.05577848937577902
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.922468185424805,
        "normalized": 0.9804199199769799,
        "contribution": 0.14706298799654696
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.152871608734131,
        "normalized": 0.9361753407327211,
        "contribution": 0.09361753407327211
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6a08f4b2e6956eac30a084ac94f907741758079d5cdb1b6458dcb5fe575a6a96",
    "id": "d4d696493d0be1bfc1e4da543b88",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 4.149612903594971,
    "wer": 0.0,
    "identity": 0.43087905645370483,
    "voiceclap": 0.026258813217282295,
    "fitness": 0.8248738552266416,
    "legacy_fitness": 0.8248738552266416,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8108951162393783,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2366087436676025,
        "normalized": 0.30305817864738066,
        "contribution": 0.07576454466184517
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.48657751083374,
        "normalized": 0.9534909337644679,
        "contribution": 0.1430236400646702
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.214478969573975,
        "normalized": 0.9427315651566268,
        "contribution": 0.09427315651566269
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f025da272c8cb3c1b91322aa3d12b0c151e4ee3452cf1d6a61437fa545896f81",
    "id": "d51dba1fe6ec55f6da4006201437",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.3127899169921875,
    "wer": 0.0,
    "identity": 0.5167893767356873,
    "voiceclap": 0.02942894771695137,
    "fitness": 0.8344208129735048,
    "legacy_fitness": 0.8344208129735048,
    "target_cdf": 0.9932207044941694,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7762160301208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7794461161607291,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7762160301208496,
        "normalized": 0.9932207044941694,
        "contribution": 0.4966103522470847
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.10091233253479,
        "normalized": 0.2508316293237907,
        "contribution": 0.06270790733094768
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.112957000732422,
        "normalized": 0.9117423574294996,
        "contribution": 0.13676135361442493
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5279998779296875,
        "normalized": 0.8336650296827179,
        "contribution": 0.0833665029682718
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "18ed692815bbf942569ebdda5458dc0cbcd4901c3a87473c73d8812207513034",
    "id": "d5470ac156c6bbefa39efa78f77b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.532451152801514,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0018299287185072899,
    "fitness": 0.7951494562928467,
    "legacy_fitness": 0.7951494562928467,
    "target_cdf": 0.8999470925936282,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5425496101379395,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6962731007369813,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5425496101379395,
        "normalized": 0.8999470925936282,
        "contribution": 0.4499735462968141
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2637479305267334,
        "normalized": 0.3140598486863819,
        "contribution": 0.07851496217159548
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.192249059677124,
        "normalized": 0.7064243659623586,
        "contribution": 0.10596365489435379
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7755134105682373,
        "normalized": 0.6182093737421804,
        "contribution": 0.061820937374218045
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "854bbebeb9a2a0564b302c86543767b738dba6af577fa485fe509cf1cfa8d7a3",
    "id": "d5ab5076ede8f7c4d717198e8255",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.292664527893066,
    "genuineness": 4.042920112609863,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.01701262779533863,
    "fitness": 0.8249444427083479,
    "legacy_fitness": 0.8249444427083479,
    "target_cdf": 0.9983456496378873,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.263189792633057,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7985667293857615,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.263189792633057,
        "normalized": 0.9983456496378873,
        "contribution": 0.49917282481894365
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.066988229751587,
        "normalized": 0.2385614708758641,
        "contribution": 0.059640367718966025
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.922468185424805,
        "normalized": 0.9804199199769799,
        "contribution": 0.14706298799654696
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.073662757873535,
        "normalized": 0.926905488513048,
        "contribution": 0.0926905488513048
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f689ff835b55f14b9b11820f6cfc920bf837f1b5b396dce4a91ad25ed7b255de",
    "id": "d60fc5ab80c3a33dbe807c2a6ff2",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.009552214294672012,
    "fitness": 0.8215339617057867,
    "legacy_fitness": 0.8215339617057867,
    "target_cdf": 0.9937903356798943,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.808681011199951,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7404897420209122,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.808681011199951,
        "normalized": 0.9937903356798943,
        "contribution": 0.49689516783994714
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5649117231369019,
        "normalized": 0.0986881865760929,
        "contribution": 0.024672046644023227
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.3086628913879395,
        "normalized": 0.9361753524593768,
        "contribution": 0.1404263028689065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3255765438079834,
        "normalized": 0.7849622466803536,
        "contribution": 0.07849622466803537
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3966d068749ec8a3d515132d61a7381a3e000d68e239b384bef1cbe9fcd9fe77",
    "id": "d6907fdc4b962c35ddb8cbc79aee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": 0.6541451215744019,
    "voiceclap": -0.07722785323858261,
    "fitness": 0.8363961470527166,
    "legacy_fitness": 0.8363961470527166,
    "target_cdf": 0.9818588758315074,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3866372108459473,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7388636892366738,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3866372108459473,
        "normalized": 0.9818588758315074,
        "contribution": 0.4909294379157537
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8600512742996216,
        "normalized": 0.17125467685227086,
        "contribution": 0.042813669213067715
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9617295265197754,
        "normalized": 0.8885304554003988,
        "contribution": 0.1332795683100598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.087949275970459,
        "normalized": 0.7184101379779255,
        "contribution": 0.07184101379779255
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "208e3717ec7398fd5c5830fdbb0f91bfc5cca9dc4bcac46d346a38da6a831181",
    "id": "d70f0acc03be022f7e1e06f67d6e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": 0.5448323488235474,
    "voiceclap": 0.03899227827787399,
    "fitness": 0.8328489066127678,
    "legacy_fitness": 0.8328489066127678,
    "target_cdf": 0.9904545229601343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6463565826416016,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7945966715506249,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6463565826416016,
        "normalized": 0.9904545229601343,
        "contribution": 0.4952272614800671
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2298238277435303,
        "normalized": 0.3003348835264067,
        "contribution": 0.07508372088160167
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.139644145965576,
        "normalized": 0.9154342857630313,
        "contribution": 0.1373151428644547
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7040200233459473,
        "normalized": 0.8697054632450133,
        "contribution": 0.08697054632450134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6d5b997625775741cec8f300826fc063c71bd5b9293ed22737f5c88d18db17be",
    "id": "d75ecfb44756182442bd82669c5d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09925692528486252,
    "fitness": 0.8351026068793004,
    "legacy_fitness": 0.8351026068793004,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.733596182474894,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9041526317596436,
        "normalized": 0.18447617999078308,
        "contribution": 0.04611904499769577
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.71264910697937,
        "normalized": 0.8413447460685429,
        "contribution": 0.12620171191028143
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1011507511138916,
        "normalized": 0.7223522856539355,
        "contribution": 0.07223522856539355
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9b1d51501af1895b4f5e6c764973f1ad1d85b096a22a61d737c443f6418fc5c8",
    "id": "d7bbdfcf55829b85c47b80e92b61",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.157639026641846,
    "genuineness": 4.005263805389404,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0643923431634903,
    "fitness": 0.8158506026454303,
    "legacy_fitness": 0.8158506026454303,
    "target_cdf": 0.9900472701763179,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.630124092102051,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7982122474269556,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.630124092102051,
        "normalized": 0.9900472701763179,
        "contribution": 0.49502363508815894
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1755452156066895,
        "normalized": 0.27895870121176836,
        "contribution": 0.06973967530294209
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.246392726898193,
        "normalized": 0.9290496824330181,
        "contribution": 0.1393574523649527
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.196876525878906,
        "normalized": 0.9409148467090185,
        "contribution": 0.09409148467090185
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ed3a9e0a4174d540403172153fe2d1058eeeba7f6738bec08b89b92a7aa1e8d0",
    "id": "d8305a73d93701b21c2baef325cb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.9299514293670654,
    "wer": 0.0,
    "identity": 0.6167869567871094,
    "voiceclap": 0.001955963671207428,
    "fitness": 0.7916395292653814,
    "legacy_fitness": 0.7916395292653814,
    "target_cdf": 0.9351928961225698,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.786036252975464,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6976940176431325,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.786036252975464,
        "normalized": 0.9351928961225698,
        "contribution": 0.4675964480612849
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.574765205383301,
        "normalized": 0.8102739199132684,
        "contribution": 0.12154108798699025
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.021941661834717,
        "normalized": 0.6983048327926806,
        "contribution": 0.06983048327926807
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a65555e59047951052bed2aca894aeb885523a6ff7303a44ba5f39bd92f4415e",
    "id": "d84697d396721ab3af3ec7f99e5b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 6.197352409362793,
    "genuineness": 3.512594699859619,
    "wer": 0.0,
    "identity": 0.6714757680892944,
    "voiceclap": 0.05800705403089523,
    "fitness": 0.7784151161032571,
    "legacy_fitness": 0.7784151161032571,
    "target_cdf": 0.9542463867500731,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9645934104919434,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7468158591086185,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9645934104919434,
        "normalized": 0.9542463867500731,
        "contribution": 0.47712319337503656
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9855704307556152,
        "normalized": 0.2105000137670302,
        "contribution": 0.05262500344175755
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9261465072631836,
        "normalized": 0.8824847562300842,
        "contribution": 0.13237271343451262
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.589607000350952,
        "normalized": 0.8469494885731179,
        "contribution": 0.0846949488573118
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4d4c8072fa4f5cf3e258d2694790e0273d9952efa072888cf99cdd9cecb7c46e",
    "id": "d85cc0523c569c39f586c5376d11",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 3.961331605911255,
    "wer": 0.0,
    "identity": 0.417279988527298,
    "voiceclap": 0.023219909518957138,
    "fitness": 0.8245739487046524,
    "legacy_fitness": 0.8245739487046524,
    "target_cdf": 0.9988438302744477,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.376816749572754,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7835637069396556,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.376816749572754,
        "normalized": 0.9988438302744477,
        "contribution": 0.49942191513722384
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9448615312576294,
        "normalized": 0.19722848067005766,
        "contribution": 0.049307120167514415
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.548847675323486,
        "normalized": 0.9585742212718315,
        "contribution": 0.1437861331907747
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.950448513031006,
        "normalized": 0.9104853844414267,
        "contribution": 0.09104853844414268
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0e2a471bba03c68ca996bd3cfda2847c3fe609edd2cccc93660f3045da6b16ec",
    "id": "d90be0de421795268007dea15289",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 6.078212261199951,
    "genuineness": 4.074300289154053,
    "wer": 0.0,
    "identity": 0.17225749790668488,
    "voiceclap": -0.0360686257481575,
    "fitness": 0.8180205142158944,
    "legacy_fitness": 0.8180205142158944,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.789053299393515,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.009317398071289,
        "normalized": 0.21847802587457404,
        "contribution": 0.05461950646864351
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.70897102355957,
        "normalized": 0.9696036562383776,
        "contribution": 0.14544054843575663
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.073662757873535,
        "normalized": 0.926905488513048,
        "contribution": 0.0926905488513048
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "54ced5b2797e56d5605e98c8d8c4df94c7cd233f678e2902eca3723018549f7e",
    "id": "d944b9f53423b377144910b06a93",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": 0.7225344181060791,
    "voiceclap": -0.026308048516511917,
    "fitness": 0.7817361281289831,
    "legacy_fitness": 0.7817361281289831,
    "target_cdf": 0.9351928961225698,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.786036252975464,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7509876268201522,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.786036252975464,
        "normalized": 0.9351928961225698,
        "contribution": 0.4675964480612849
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1110897064208984,
        "normalized": 0.25457682692833283,
        "contribution": 0.06364420673208321
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9617295265197754,
        "normalized": 0.8885304554003988,
        "contribution": 0.1332795683100598
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.677617073059082,
        "normalized": 0.8646740371672423,
        "contribution": 0.08646740371672423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "31c6e2eae6236ba4f5b080871aef7924336a0db3369f2779efd637b40e6fa4a4",
    "id": "d9eb1d27f571d1e38e31fdde4aee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": 0.5792437195777893,
    "voiceclap": -0.10332818329334259,
    "fitness": 0.8341861832971542,
    "legacy_fitness": 0.8341861832971542,
    "target_cdf": 0.9763929662774343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.273009777069092,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7436049595774205,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.273009777069092,
        "normalized": 0.9763929662774343,
        "contribution": 0.48819648313871716
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9991400241851807,
        "normalized": 0.2150377207338816,
        "contribution": 0.0537594301834704
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7215447425842285,
        "normalized": 0.8432277384318043,
        "contribution": 0.12648416076477065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.202362537384033,
        "normalized": 0.7516488549046221,
        "contribution": 0.07516488549046221
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1f9d1a77f89796dc16e11f19e04c6795027ba9f1eff9fa67e7935fd7241cd99e",
    "id": "d9ffa1e904f11d0a25a0e9dbb63d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.483288764953613,
    "genuineness": 4.030367851257324,
    "wer": 0.0,
    "identity": 0.7449097037315369,
    "voiceclap": -0.02119695395231247,
    "fitness": 0.7665076947782321,
    "legacy_fitness": 0.7665076947782321,
    "target_cdf": 0.8729869637040064,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3964571952819824,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6466672657922612,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3964571952819824,
        "normalized": 0.8729869637040064,
        "contribution": 0.4364934818520032
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8023803234100342,
        "normalized": 0.1549039932623571,
        "contribution": 0.038725998315589275
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2678627967834473,
        "normalized": 0.7288620844464377,
        "contribution": 0.10932931266696566
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7843143939971924,
        "normalized": 0.6211847295770306,
        "contribution": 0.062118472957703064
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "871ad6dbe972175d8d3ffceabbb2327ac19fc24bf58a67958882c255f5e45997",
    "id": "da4e072eaa81675508f13fa92a81",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.316492557525635,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": 0.5659171342849731,
    "voiceclap": -0.12189708650112152,
    "fitness": 0.8114688545300317,
    "legacy_fitness": 0.8114688545300317,
    "target_cdf": 0.9409148593413338,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.834733486175537,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7334597606146702,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.834733486175537,
        "normalized": 0.9409148593413338,
        "contribution": 0.4704574296706669
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1484060287475586,
        "normalized": 0.2685555073584547,
        "contribution": 0.06713887683961367
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4991517066955566,
        "normalized": 0.7917476292303571,
        "contribution": 0.11876214438455357
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.272770404815674,
        "normalized": 0.7710130971983604,
        "contribution": 0.07710130971983604
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "87ef475b1d466a915ab74dd2733355470219cbbca1448b94525c5232dbd53626",
    "id": "da6507cc8c5f8db8ff6032a19633",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": 0.43736690282821655,
    "voiceclap": -0.0596308708190918,
    "fitness": 0.7996481085838363,
    "legacy_fitness": 0.7996481085838363,
    "target_cdf": 0.9619218677325021,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0538716316223145,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7083986196624652,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0538716316223145,
        "normalized": 0.9619218677325021,
        "contribution": 0.48096093386625105
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8638763427734375,
        "normalized": 0.8713534713164917,
        "contribution": 0.13070302069747375
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1627578735351562,
        "normalized": 0.740384500634085,
        "contribution": 0.0740384500634085
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9d3fc975bad73c001133f3275e8319eeb44b212847ecb89815ba9f69cb702745",
    "id": "da7343ba6de7c4663c65a167a5ca",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.011797893792390823,
    "fitness": 0.818726369400253,
    "legacy_fitness": 0.818726369400253,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.700974088017558,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.2595947980880737,
        "normalized": 0.050437634734188164,
        "contribution": 0.012609408683547041
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.979520797729492,
        "normalized": 0.8914683718087433,
        "contribution": 0.1337202557713115
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7755134105682373,
        "normalized": 0.6182093737421804,
        "contribution": 0.061820937374218045
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "72670c528f01361f1d8f53e7d46d10c8c37132e2bf532cce73b8c4afb3f7cabe",
    "id": "da889cb0f1202ec509d06cc51745",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.23706579208374,
    "genuineness": 3.5816311836242676,
    "wer": 0.0,
    "identity": 0.7030646800994873,
    "voiceclap": 0.0296986922621727,
    "fitness": 0.8024439697159645,
    "legacy_fitness": 0.8024439697159645,
    "target_cdf": 0.9945653845257199,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8573784828186035,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7928455540063163,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8573784828186035,
        "normalized": 0.9945653845257199,
        "contribution": 0.4972826922628599
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0262794494628906,
        "normalized": 0.2242816522895154,
        "contribution": 0.05607041307237885
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.851302146911621,
        "normalized": 0.9772498567468411,
        "contribution": 0.14658747851202616
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.091264724731445,
        "normalized": 0.9290497015905134,
        "contribution": 0.09290497015905134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c3b796154580a24086bee6c159ca8344f28d2f8577bdaa8cebad7febc365e085",
    "id": "daba6d7d45ddac83ebf48216262e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1,
    "generation": null,
    "ce": 6.149696350097656,
    "genuineness": 4.557555198669434,
    "wer": 0.0,
    "identity": 0.20952442288398743,
    "voiceclap": -0.10646331310272217,
    "fitness": 0.6774946784599846,
    "legacy_fitness": 0.6774946784599846,
    "target_cdf": 0.6677082728593002,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.6619387865066528,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.5669625599667167,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.6619387865066528,
        "normalized": 0.6677082728593002,
        "contribution": 0.3338541364296501
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4588115215301514,
        "normalized": 0.39752227638752075,
        "contribution": 0.09938056909688019
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.605130910873413,
        "normalized": 0.510907229866791,
        "contribution": 0.07663608448001866
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.637997627258301,
        "normalized": 0.5709176996016775,
        "contribution": 0.05709176996016775
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4534ebb63e6afef7d69113070f3f00f350bf9b7b62592feca07ee1ff04e1fb12",
    "id": "dabb07629fa9dcc49ba65226cfd6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 3.7793264389038086,
    "wer": 0.0,
    "identity": 0.5971173048019409,
    "voiceclap": 0.07125857472419739,
    "fitness": 0.8150549604729282,
    "legacy_fitness": 0.8150549604729282,
    "target_cdf": 0.9997385569881334,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.815093040466309,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8443119088339262,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.815093040466309,
        "normalized": 0.9997385569881334,
        "contribution": 0.4998692784940667
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.436760902404785,
        "normalized": 0.3877622289930795,
        "contribution": 0.09694055724826987
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.634126663208008,
        "normalized": 0.9964005462960434,
        "contribution": 0.1494600819444065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.760141372680664,
        "normalized": 0.9804199114718302,
        "contribution": 0.09804199114718302
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e3e91f17bc4f435f2127ae4d43a6e20c274526e6c59e1ef447b037f4ef240ccb",
    "id": "dae77858fcfead83a2be4395b4ff",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.600459337234497,
    "wer": 0.0,
    "identity": 0.45384481549263,
    "voiceclap": 0.034836724400520325,
    "fitness": 0.7985385479391669,
    "legacy_fitness": 0.7985385479391669,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7370846482155657,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6191902160644531,
        "normalized": 0.1099935309435226,
        "contribution": 0.02749838273588065
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.3086628913879395,
        "normalized": 0.9361753524593768,
        "contribution": 0.1404263028689065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3431787490844727,
        "normalized": 0.7895000148839948,
        "contribution": 0.07895000148839948
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "31346aecb0f8e7b522b51fcbd07bbd47826516b68946f15dfc3a04ab692e78f2",
    "id": "dafce1d08689f365e148afabbd2f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.562715530395508,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.13092041015625,
    "fitness": 0.8240082484031152,
    "legacy_fitness": 0.8240082484031152,
    "target_cdf": 0.9662346008508267,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.110685348510742,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7166156915845119,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.110685348510742,
        "normalized": 0.9662346008508267,
        "contribution": 0.48311730042541334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5920509099960327,
        "normalized": 0.10422977602228944,
        "contribution": 0.02605744400557236
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.215564012527466,
        "normalized": 0.7553448800736362,
        "contribution": 0.07553448800736362
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f080098fd7b971f83a32313f0817036179ca2d4d7d3cff73e6175d70ef7f8a25",
    "id": "db2dc1cd3f699b9f7a71325c01df",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": 0.6537904143333435,
    "voiceclap": -0.13788221776485443,
    "fitness": 0.8147760153774162,
    "legacy_fitness": 0.8147760153774162,
    "target_cdf": 0.9904545229601343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6463565826416016,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7289310032290195,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6463565826416016,
        "normalized": 0.9904545229601343,
        "contribution": 0.4952272614800671
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.828293561935425,
        "normalized": 0.8646740370257361,
        "contribution": 0.1297011055538604
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.087949275970459,
        "normalized": 0.7184101379779255,
        "contribution": 0.07184101379779255
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0f9f6ca803b0a20a8d2b5a70f9e640f925c31e2f4cb4fbe2d4f7fb74fc5ff0af",
    "id": "db5bfd494fa0d3a6f744adbe244e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10637306421995163,
    "fitness": 0.8334361061586992,
    "legacy_fitness": 0.8334361061586992,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.790039675303225,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.066988229751587,
        "normalized": 0.2385614708758641,
        "contribution": 0.059640367718966025
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.442098617553711,
        "normalized": 0.9495623364922869,
        "contribution": 0.14243435047384304
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.888841390609741,
        "normalized": 0.9013118211321569,
        "contribution": 0.09013118211321569
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "37b6e65deb47f6d187ce54d8f439a8554499815a11fa12327a8a0395efbae945",
    "id": "dc28119a2fd620f8687c5032b079",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.483288764953613,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12265459448099136,
    "fitness": 0.8324300977409583,
    "legacy_fitness": 0.8324300977409583,
    "target_cdf": 0.9922796672226393,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7275185585021973,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.701061893369108,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7275185585021973,
        "normalized": 0.9922796672226393,
        "contribution": 0.49613983361131964
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.1238985061645508,
        "normalized": 0.03617773562425208,
        "contribution": 0.00904443390606302
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.255288600921631,
        "normalized": 0.9301035297844544,
        "contribution": 0.13951552946766815
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6170949935913086,
        "normalized": 0.5636209638405715,
        "contribution": 0.05636209638405715
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c8fd3406c4f15ef570155ef5163eeffdf454f3be3ed6b69de92adc8814ac897d",
    "id": "dc6c01b4e5a9275c9ffb013e68f5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.181467056274414,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": 0.3223903477191925,
    "voiceclap": -0.12388473749160767,
    "fitness": 0.8047692998139044,
    "legacy_fitness": 0.8047692998139044,
    "target_cdf": 0.9321750211493136,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.761687755584717,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7392243147508015,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.761687755584717,
        "normalized": 0.9321750211493136,
        "contribution": 0.4660875105746568
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1280517578125,
        "normalized": 0.2608832234596053,
        "contribution": 0.06522080586490132
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.685961961746216,
        "normalized": 0.835607112993103,
        "contribution": 0.12534106694896543
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.492795944213867,
        "normalized": 0.8257493136227798,
        "contribution": 0.08257493136227799
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b90431dc8c3b10cf65646b823ca29d40faa6b5d07ea5a3734db4ae1c07395abc",
    "id": "dc90aaa50b007065f5558e8722a7",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": 0.6339399814605713,
    "voiceclap": -0.0059216879308223724,
    "fitness": 0.8390321000090775,
    "legacy_fitness": 0.8390321000090775,
    "target_cdf": 0.9987822818103504,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.360584259033203,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7363443804547105,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.360584259033203,
        "normalized": 0.9987822818103504,
        "contribution": 0.4993911409051752
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5309876203536987,
        "normalized": 0.09206841268091792,
        "contribution": 0.02301710317022948
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.29976749420166,
        "normalized": 0.9351929313530816,
        "contribution": 0.14027893970296224
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1495563983917236,
        "normalized": 0.7365719667634352,
        "contribution": 0.07365719667634352
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c785b36f795d72ab1df2e10f90c2bd677e3bdf169c8750028cc3ef5bd333739e",
    "id": "dcb16942501490cde6e09d62182d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.325342178344727,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.05486871302127838,
    "fitness": 0.8392648118858382,
    "legacy_fitness": 0.8392648118858382,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7259906130498017,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5513420104980469,
        "normalized": 0.09599963185428989,
        "contribution": 0.023999907963572473
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.1485395431518555,
        "normalized": 0.9166387807822887,
        "contribution": 0.1374958171173433
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.942732572555542,
        "normalized": 0.67336307136718,
        "contribution": 0.067336307136718
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "333dd2ff33a38e1952aec360393321b24b7b4864e1a8f6fff2bf4aa96a5b9cb1",
    "id": "dd089763aa6c083f91cb767f10f1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 3.7542223930358887,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0029906854033470154,
    "fitness": 0.8117320820094991,
    "legacy_fitness": 0.8117320820094991,
    "target_cdf": 0.9896249303026692,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6138916015625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.726648577188231,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6138916015625,
        "normalized": 0.9896249303026692,
        "contribution": 0.4948124651513346
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.4834939241409302,
        "normalized": 0.0833611930352981,
        "contribution": 0.020840298258824524
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.1663312911987305,
        "normalized": 0.9190091380825917,
        "contribution": 0.13785137071238873
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1319541931152344,
        "normalized": 0.7314444306568307,
        "contribution": 0.07314444306568307
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b65a0ef8d61527b7df9413a9de1be17cbab29ff1674050a045894b9fc2381127",
    "id": "dd10797d0147e9372e06bf49d200",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.406930446624756,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.05740112438797951,
    "fitness": 0.7639577400555855,
    "legacy_fitness": 0.7639577400555855,
    "target_cdf": 0.8384925309559991,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2381908893585205,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.62461346768686,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2381908893585205,
        "normalized": 0.8384925309559991,
        "contribution": 0.41924626547799954
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1076972484588623,
        "normalized": 0.25332518159841916,
        "contribution": 0.06333129539960479
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.722999334335327,
        "normalized": 0.5520567409912419,
        "contribution": 0.08280851114868629
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6996045112609863,
        "normalized": 0.5922739566056935,
        "contribution": 0.05922739566056935
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ac51efefbd2bd80a73dadd8566c78397af675f41917187784bc62e7badaaa319",
    "id": "dd75e8408a959b93860807717365",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.026641840115189552,
    "fitness": 0.8205128269638393,
    "legacy_fitness": 0.8205128269638393,
    "target_cdf": 0.996049248624995,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.971005439758301,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8550314495877142,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.971005439758301,
        "normalized": 0.996049248624995,
        "contribution": 0.4980246243124975
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.575849771499634,
        "normalized": 0.4502618261088747,
        "contribution": 0.11256545652721868
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.993634223937988,
        "normalized": 0.9832067001034072,
        "contribution": 0.14748100501551106
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.548917293548584,
        "normalized": 0.9696036373248687,
        "contribution": 0.09696036373248687
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "639f434e1c68ad8629fe7ac225b240df0ca54a7ea87aafaf186a0d787a7b374a",
    "id": "ddd6c55e1725ce488617b7d444b1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 3.9425034523010254,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07592273503541946,
    "fitness": 0.7578303047619519,
    "legacy_fitness": 0.7578303047619519,
    "target_cdf": 0.8697054919803655,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3802249431610107,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6126273884535764,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3802249431610107,
        "normalized": 0.8697054919803655,
        "contribution": 0.43485274599018275
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6463295221328735,
        "normalized": 0.11598248129215816,
        "contribution": 0.02899562032303954
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.005438804626465,
        "normalized": 0.6476213017017056,
        "contribution": 0.09714319525525585
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.482879638671875,
        "normalized": 0.5163582688509835,
        "contribution": 0.05163582688509835
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "43ff71d5fb793c13d0d810b18aa94db5368dbde07910c8211174e9db85138567",
    "id": "de2d1d24d248526985741f3462f0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 4.595211505889893,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": -0.16199184954166412,
    "fitness": 0.7637256732708669,
    "legacy_fitness": 0.7637256732708669,
    "target_cdf": 0.8824847618116329,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4451546669006348,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6195432497591143,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4451546669006348,
        "normalized": 0.8824847618116329,
        "contribution": 0.44124238090581647
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7481017112731934,
        "normalized": 0.14048896779652476,
        "contribution": 0.03512224194913119
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 2.8152923583984375,
        "normalized": 0.5839129218252099,
        "contribution": 0.08758693827378149
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.595092535018921,
        "normalized": 0.5559168863038523,
        "contribution": 0.05559168863038523
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4556ca6b1e9325840aca4533f5d02aa9a81a69347393cbc1904923c49e9307ef",
    "id": "de465083f4a48a88fe0e781797b8",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.036643981933594,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06434014439582825,
    "fitness": 0.8243720538158998,
    "legacy_fitness": 0.8243720538158998,
    "target_cdf": 0.9862070255954634,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5002641677856445,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7319652877150215,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5002641677856445,
        "normalized": 0.9862070255954634,
        "contribution": 0.4931035127977317
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6056206226348877,
        "normalized": 0.10708370011583324,
        "contribution": 0.02677092502895831
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.086269855499268,
        "normalized": 0.9079316126330063,
        "contribution": 0.13618974189495092
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2287654876708984,
        "normalized": 0.7590110799338052,
        "contribution": 0.07590110799338053
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3ab94b5f2800c8375b4a3a73cb6e53e50b63b17efad1bd68dd7fac15ee31adc9",
    "id": "de7372dcd7845c3b2528bf325195",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.602428436279297,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.24830889701843262,
    "fitness": 0.769139815559926,
    "legacy_fitness": 0.769139815559926,
    "target_cdf": 0.834637954217081,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.221958637237549,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6036456299956087,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.221958637237549,
        "normalized": 0.834637954217081,
        "contribution": 0.4173189771085405
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6802536249160767,
        "normalized": 0.12378946157536314,
        "contribution": 0.030947365393840784
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.192249059677124,
        "normalized": 0.7064243659623586,
        "contribution": 0.10596365489435379
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.4201724529266357,
        "normalized": 0.49415632598873566,
        "contribution": 0.04941563259887357
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7d4d5bcecef1a34a63b6d3e08e4189f42675472fa1626237859a3ac216728b65",
    "id": "dfbc443a963e1025a8d524c88253",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.231201171875,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.1855967938899994,
    "fitness": 0.7466869280314343,
    "legacy_fitness": 0.7466869280314343,
    "target_cdf": 0.9351928961225698,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.786036252975464,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7175623593183382,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.786036252975464,
        "normalized": 0.9351928961225698,
        "contribution": 0.4675964480612849
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9923553466796875,
        "normalized": 0.2127618447245091,
        "contribution": 0.053190461181127274
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.601452350616455,
        "normalized": 0.8165628197935635,
        "contribution": 0.12248442296903451
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1715588569641113,
        "normalized": 0.7429102710689157,
        "contribution": 0.07429102710689157
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "43afb614d0321de0cadfe322515bafe63d56f7f20528e74feb9d730f06ea8bd0",
    "id": "e003484ece1ef78f8fa4677a0c77",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.165581703186035,
    "genuineness": 4.250029563903809,
    "wer": 0.06666666666666667,
    "identity": null,
    "voiceclap": -0.10803551971912384,
    "fitness": 0.7451019096788801,
    "legacy_fitness": 0.7451019096788801,
    "target_cdf": 0.9341987338272668,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7779200077056885,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7543450660132893,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7779200077056885,
        "normalized": 0.9341987338272668,
        "contribution": 0.4670993669136334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1755452156066895,
        "normalized": 0.27895870121176836,
        "contribution": 0.06973967530294209
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.908355236053467,
        "normalized": 0.8793763943077506,
        "contribution": 0.13190645914616259
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6336121559143066,
        "normalized": 0.8559956465055119,
        "contribution": 0.08559956465055119
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a350f45d5cf0724c1ddb24d66b961ef02d6c97f40ca840cc3fe3694eaa05e0ec",
    "id": "e01e99e33e29467904464d45be2c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.229123115539551,
    "genuineness": 3.562803030014038,
    "wer": 0.0,
    "identity": 0.6217362284660339,
    "voiceclap": 0.12192870676517487,
    "fitness": 0.8009337278039066,
    "legacy_fitness": 0.8009337278039066,
    "target_cdf": 0.9935110473296658,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7924485206604004,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.799035887029772,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7924485206604004,
        "normalized": 0.9935110473296658,
        "contribution": 0.4967555236648329
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1144819259643555,
        "normalized": 0.25583161582232095,
        "contribution": 0.06395790395558024
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.7979278564453125,
        "normalized": 0.974597599214758,
        "contribution": 0.1461896398822137
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.029657363891602,
        "normalized": 0.921328195271452,
        "contribution": 0.09213281952714521
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "067cabe6f209e3f3455b8f0766a964baf8ace8cc7e30c758d98eee19a4b8e17d",
    "id": "e023e2135a086f838e5d5ef6ae9c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.218649387359619,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.015463149175047874,
    "fitness": 0.8278902417871796,
    "legacy_fitness": 0.8278902417871796,
    "target_cdf": 0.9796648197280255,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.337939739227295,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7210314497703674,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.337939739227295,
        "normalized": 0.9796648197280255,
        "contribution": 0.4898324098640128
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5716965198516846,
        "normalized": 0.1000529371132764,
        "contribution": 0.0250132342783191
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.139644145965576,
        "normalized": 0.9154342857630313,
        "contribution": 0.1373151428644547
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.991138219833374,
        "normalized": 0.688706627635808,
        "contribution": 0.06887066276358081
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a5426043747c2072e7ee8d12044f48a2bb1c021c6a9e789c120f03a7d7eb8888",
    "id": "e06f6730ce9cc984de2173fdcf1e",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.193545341491699,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.14713707566261292,
    "fitness": 0.7941332722344712,
    "legacy_fitness": 0.7941332722344712,
    "target_cdf": 0.9154342620847058,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.639944076538086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6629429792400608,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.639944076538086,
        "normalized": 0.9154342620847058,
        "contribution": 0.4577171310423529
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4012985229492188,
        "normalized": 0.7662529752285923,
        "contribution": 0.11493794628428883
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7711129188537598,
        "normalized": 0.6167190605492896,
        "contribution": 0.06167190605492896
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d45e1328cb2cbd6f3669ec2a2b175ab96ba38904ed55de0042f34a6f86abf020",
    "id": "e0e9cb56020a47d036c664b14af4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 3.747946262359619,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.116380974650383,
    "fitness": 0.7898109710499734,
    "legacy_fitness": 0.7898109710499734,
    "target_cdf": 0.9427315386702796,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.850965976715088,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7435480435498535,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.850965976715088,
        "normalized": 0.9427315386702796,
        "contribution": 0.4713657693351398
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.16536808013916,
        "normalized": 0.2750346407474058,
        "contribution": 0.06875866018685145
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.539182424545288,
        "normalized": 0.8016852878192227,
        "contribution": 0.1202527931728834
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5191988945007324,
        "normalized": 0.8317082085497887,
        "contribution": 0.08317082085497887
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0a8df4265f2a6ae75f204c09e5502c71ffc5ef86fb35100ac5692e3b3cd437fe",
    "id": "e15532fd38ba550cff75cd0260df",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 5.895530700683594,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07290733605623245,
    "fitness": 0.8147609825589933,
    "legacy_fitness": 0.8147609825589933,
    "target_cdf": 0.9926053912756201,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.743751049041748,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8377103724938482,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.743751049041748,
        "normalized": 0.9926053912756201,
        "contribution": 0.49630269563781004
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5054571628570557,
        "normalized": 0.4183740949029872,
        "contribution": 0.1045935237257468
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.477681636810303,
        "normalized": 0.9527255009885056,
        "contribution": 0.14290882514827583
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.179274559020996,
        "normalized": 0.939053279820155,
        "contribution": 0.09390532798201551
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "353122760e143e6fc7c783a5e52bc91413a26d759da3f697e5962c0605a80b45",
    "id": "e16a5824f276a84deb9c2ff0b0fb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": 0.6648473739624023,
    "voiceclap": -0.0032293852418661118,
    "fitness": 0.7823911763634227,
    "legacy_fitness": 0.7823911763634227,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6784748536063645,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6191902160644531,
        "normalized": 0.1099935309435226,
        "contribution": 0.02749838273588065
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6236917972564697,
        "normalized": 0.8217031044875092,
        "contribution": 0.12325546567312637
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9163296222686768,
        "normalized": 0.6648664054810712,
        "contribution": 0.06648664054810713
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5112e25872752cd9593070948070255d440cbecdd978ab7f9f47d03b44c184ad",
    "id": "e1c5a9db32440dbfce9b6f3891d1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 3,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 3.8922953605651855,
    "wer": 0.0,
    "identity": 0.6977342367172241,
    "voiceclap": 0.051650337874889374,
    "fitness": 0.806608106194704,
    "legacy_fitness": 0.806608106194704,
    "target_cdf": 0.9690622995272662,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.151266574859619,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.796619457103417,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.151266574859619,
        "normalized": 0.9690622995272662,
        "contribution": 0.4845311497636331
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3095455169677734,
        "normalized": 0.3330013083601581,
        "contribution": 0.08325032709003953
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.210809707641602,
        "normalized": 0.9247119875768577,
        "contribution": 0.13870679813652864
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.888841390609741,
        "normalized": 0.9013118211321569,
        "contribution": 0.09013118211321569
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3d7ff02c7f0698eec0e151db43cb693fc507e80e833de94229bafbfc35e66a59",
    "id": "e20e8fce6951e7d5bf8352ad4e9b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.213237762451172,
    "genuineness": 4.218649387359619,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.05806529521942139,
    "fitness": 0.8293907293638895,
    "legacy_fitness": 0.8293907293638895,
    "target_cdf": 0.9965654549919187,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.019702911376953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7545896769185062,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.019702911376953,
        "normalized": 0.9965654549919187,
        "contribution": 0.49828272749595937
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5920509099960327,
        "normalized": 0.10422977602228944,
        "contribution": 0.02605744400557236
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.433203220367432,
        "normalized": 0.9487457892743885,
        "contribution": 0.14231186839115825
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.756826162338257,
        "normalized": 0.8793763702581625,
        "contribution": 0.08793763702581625
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b4300ad69618b62c5b367c403fa580f5f61f0acbf1c55099a3e6ca5f32adaf9e",
    "id": "e21333cbd85734891de3d11b2660",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.218649387359619,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12209343910217285,
    "fitness": 0.828101393242111,
    "legacy_fitness": 0.828101393242111,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7436826049991728,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.157435417175293,
        "normalized": 0.9178303974842037,
        "contribution": 0.13767455962263056
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3255765438079834,
        "normalized": 0.7849622466803536,
        "contribution": 0.07849622466803537
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "22d448d35a38e64a098aaa5578c47f9e6274e1384dffa46d9c5ff8a81a388dd4",
    "id": "e27df83d4c87c6c8022e42f4b4ba",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": 0.5610034465789795,
    "voiceclap": -0.01798165775835514,
    "fitness": 0.8198889691739573,
    "legacy_fitness": 0.8198889691739573,
    "target_cdf": 0.9736578860102679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2243127822875977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7270454905795637,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2243127822875977,
        "normalized": 0.9736578860102679,
        "contribution": 0.48682894300513396
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.032895088195801,
        "normalized": 0.8999470264638179,
        "contribution": 0.13499205396957267
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1011507511138916,
        "normalized": 0.7223522856539355,
        "contribution": 0.07223522856539355
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "23d0c00b7eb5af3f1ed212f65eb02974613ba95599f7c6b2b98137c4baf2235f",
    "id": "e2dcb007663c769fddabe2cbd559",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0918443351984024,
    "fitness": 0.775821883576403,
    "legacy_fitness": 0.775821883576403,
    "target_cdf": 0.8663657806263168,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.36399245262146,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6555403883629706,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.36399245262146,
        "normalized": 0.8663657806263168,
        "contribution": 0.4331828903131584
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9245071411132812,
        "normalized": 0.1907869738584032,
        "contribution": 0.0476967434646008
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1410984992980957,
        "normalized": 0.6907744870138126,
        "contribution": 0.1036161730520719
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0615463256835938,
        "normalized": 0.710445815331395,
        "contribution": 0.0710445815331395
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "52f447d42ce2898c88e8cf8c8f74531043cc5d82e07032de0cf372fe0e466ea2",
    "id": "e34b062ea3e4ad0ccd9b3e435944",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 3,
    "generation": null,
    "ce": 5.974957466125488,
    "genuineness": 3.8295347690582275,
    "wer": 0.0,
    "identity": 0.2975880205631256,
    "voiceclap": 0.0800214558839798,
    "fitness": 0.807823234569509,
    "legacy_fitness": 0.807823234569509,
    "target_cdf": 0.9977706984112247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.165794849395752,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8407599980445271,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.165794849395752,
        "normalized": 0.9977706984112247,
        "contribution": 0.49888534920561234
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.421494960784912,
        "normalized": 0.381046116164786,
        "contribution": 0.0952615290411965
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.135965347290039,
        "normalized": 0.9877755173761718,
        "contribution": 0.14816632760642576
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.865753650665283,
        "normalized": 0.9844679219129244,
        "contribution": 0.09844679219129245
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "7c0f66e3c1ded1c78b5b53d1d22111030aaf2719448b930f128568eeef210100",
    "id": "e3615377812a3570b20b1bdca787",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09475982189178467,
    "fitness": 0.8124209459721551,
    "legacy_fitness": 0.8124209459721551,
    "target_cdf": 0.9269054842192939,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.72110652923584,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6618557767127338,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.72110652923584,
        "normalized": 0.9269054842192939,
        "contribution": 0.46345274210964693
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6395447254180908,
        "normalized": 0.11446398343396053,
        "contribution": 0.02861599585849013
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3568201065063477,
        "normalized": 0.7541161908392326,
        "contribution": 0.11311742862588489
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6258962154388428,
        "normalized": 0.5666961011871179,
        "contribution": 0.05666961011871179
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3dee491a9a5a59633b4ca872ddb2f796831f238f44e569a9345e53512c783586",
    "id": "e3c3086244e8688b5270f924342d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.53094482421875,
    "genuineness": 3.8044304847717285,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.016846448183059692,
    "fitness": 0.80370268697875,
    "legacy_fitness": 0.80370268697875,
    "target_cdf": 0.9638222590155852,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0782203674316406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8303036875052893,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0782203674316406,
        "normalized": 0.9638222590155852,
        "contribution": 0.4819111295077926
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.6148624420166016,
        "normalized": 0.4680875947828935,
        "contribution": 0.11702189869572338
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.1663312911987305,
        "normalized": 0.9190091380825917,
        "contribution": 0.13785137071238873
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.144070625305176,
        "normalized": 0.9351928858938454,
        "contribution": 0.09351928858938455
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e07c4a4f60275f89a5561cfe9e541bb022f6174f5086dcd5a336416cf1e98176",
    "id": "e40842e03315a89724eab91abb75",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.419747352600098,
    "genuineness": 4.362998008728027,
    "wer": 0.0,
    "identity": 0.7171903252601624,
    "voiceclap": 0.010580189526081085,
    "fitness": 0.8378980274068372,
    "legacy_fitness": 0.8378980274068372,
    "target_cdf": 0.9912255197896671,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.678821086883545,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7486192347726428,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.678821086883545,
        "normalized": 0.9912255197896671,
        "contribution": 0.49561275989483355
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8668360710144043,
        "normalized": 0.17324834271715545,
        "contribution": 0.04331208567928886
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.9439380168914795,
        "normalized": 0.8855360174979171,
        "contribution": 0.13283040262468757
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2639694213867188,
        "normalized": 0.7686398657383282,
        "contribution": 0.07686398657383282
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c05a9c6558c4670362a99319fe7fe628e8f81e718da88469fb5b97df9b6cc39f",
    "id": "e43d42d46474dc35420c59068384",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.118232727050781,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07519839704036713,
    "fitness": 0.8019337237555196,
    "legacy_fitness": 0.8019337237555196,
    "target_cdf": 0.9381054839846314,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.81038498878479,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7234089953439868,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.81038498878479,
        "normalized": 0.9381054839846314,
        "contribution": 0.4690527419923157
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.10091233253479,
        "normalized": 0.2508316293237907,
        "contribution": 0.06270790733094768
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4991517066955566,
        "normalized": 0.7917476292303571,
        "contribution": 0.11876214438455357
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1231532096862793,
        "normalized": 0.7288620163616988,
        "contribution": 0.07288620163616988
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e9e3015c9d9ac4ffdcb4ae89c10f78acb777c59b2f8a161c479012a327923e79",
    "id": "e4442391a6eacb870406f166b0ff",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1228010430932045,
    "fitness": 0.7600804991811385,
    "legacy_fitness": 0.7600804991811385,
    "target_cdf": 0.8506121848602832,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2909464836120605,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6643738893692764,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2909464836120605,
        "normalized": 0.8506121848602832,
        "contribution": 0.4253060924301416
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.073773145675659,
        "normalized": 0.24098894914960267,
        "contribution": 0.06024723728740067
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.321237087249756,
        "normalized": 0.744168371280294,
        "contribution": 0.1116252556920441
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9383320808410645,
        "normalized": 0.6719530395969003,
        "contribution": 0.06719530395969003
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b4d7ebce7d86ba402951796fa54c2bcc773c76ea664c655e286c924d94479b2e",
    "id": "e580893f5722f0388cc5d59ded60",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": 0.7611289024353027,
    "voiceclap": 0.04700198024511337,
    "fitness": 0.8018073163291866,
    "legacy_fitness": 0.8018073163291866,
    "target_cdf": 0.9804199222447583,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3541722297668457,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7235775583230256,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3541722297668457,
        "normalized": 0.9804199222447583,
        "contribution": 0.49020996112237913
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.415645718574524,
        "normalized": 0.07201630336001663,
        "contribution": 0.018004075840004158
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.175227165222168,
        "normalized": 0.9201750705441492,
        "contribution": 0.13802626058162237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.281571388244629,
        "normalized": 0.7733726077901995,
        "contribution": 0.07733726077901995
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "45db6ad14a72855a45e8d82b6aa22132806c935d5836eb8b9a7e64d5bd1c0890",
    "id": "e581cd2c54925f70f654f37361f5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09473916888237,
    "fitness": 0.8167191930148727,
    "legacy_fitness": 0.8167191930148727,
    "target_cdf": 0.9638222590155852,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0782203674316406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7207754958694002,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0782203674316406,
        "normalized": 0.9638222590155852,
        "contribution": 0.4819111295077926
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8600512742996216,
        "normalized": 0.17125467685227086,
        "contribution": 0.042813669213067715
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.748231887817383,
        "normalized": 0.8487881729410258,
        "contribution": 0.12731822594115386
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9867377281188965,
        "normalized": 0.6873247120738596,
        "contribution": 0.06873247120738596
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "29b594ea69c5e470c0f1859f8b3ca7a0b7d1b1407a53b19582b1520a74488f1c",
    "id": "e63d44135bb4e3284e502022ec6c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.252951145172119,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.03682628273963928,
    "fitness": 0.8268940538829828,
    "legacy_fitness": 0.8268940538829828,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7607049673922445,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9889628887176514,
        "normalized": 0.21162916044311025,
        "contribution": 0.05290729011077756
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.041790962219238,
        "normalized": 0.9013118039226871,
        "contribution": 0.13519677058840304
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5368008613586426,
        "normalized": 0.8356070969154066,
        "contribution": 0.08356070969154067
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0ac54ea931b3f91354a817f1a2b18ec6c6fac9fca776fa0014a0e14079c4019d",
    "id": "e6499d4cc493af95f3501eec5c87",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.538887023925781,
    "genuineness": 3.873466968536377,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07332609593868256,
    "fitness": 0.8171005997266485,
    "legacy_fitness": 0.8171005997266485,
    "target_cdf": 0.9844679342123099,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4515671730041504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7396931582379478,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4515671730041504,
        "normalized": 0.9844679342123099,
        "contribution": 0.49223396710615497
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6802536249160767,
        "normalized": 0.12378946157536314,
        "contribution": 0.030947365393840784
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.210809707641602,
        "normalized": 0.9247119875768577,
        "contribution": 0.13870679813652864
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.299173593521118,
        "normalized": 0.7780502760142334,
        "contribution": 0.07780502760142334
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d3428b7275b544a48284c837b672ebd1989aa827f8ca0bc5ea2159856aa0ab87",
    "id": "e6d202bf39bd549921fa62698c6c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.25018393993377686,
    "fitness": 0.7726022846991799,
    "legacy_fitness": 0.7726022846991799,
    "target_cdf": 0.8680429235090776,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3721086978912354,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6231542656042781,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3721086978912354,
        "normalized": 0.8680429235090776,
        "contribution": 0.4340214617545388
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6802536249160767,
        "normalized": 0.12378946157536314,
        "contribution": 0.030947365393840784
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1878011226654053,
        "normalized": 0.7050781385006836,
        "contribution": 0.10576172077510253
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.505157232284546,
        "normalized": 0.5242371768079602,
        "contribution": 0.052423717680796025
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "36017dedc808078ac812c267ae92c2941d18406d3eaf9e68b285ebc806fbbc37",
    "id": "e6d47111d2e39beb12322e84d670",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.7793264389038086,
    "wer": 0.0,
    "identity": 0.5184609293937683,
    "voiceclap": -0.0736156553030014,
    "fitness": 0.744970610157158,
    "legacy_fitness": 0.744970610157158,
    "target_cdf": 0.8559956519117078,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3152952194213867,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6537289094884812,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3152952194213867,
        "normalized": 0.8559956519117078,
        "contribution": 0.4279978259558539
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8498740196228027,
        "normalized": 0.16829178263767364,
        "contribution": 0.04207294565941841
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.34792423248291,
        "normalized": 0.7516488247877673,
        "contribution": 0.1127473237181651
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.057145595550537,
        "normalized": 0.7091081415504378,
        "contribution": 0.07091081415504379
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "553442aeba131e3c24d943ece8e2edc5c9beeb77a4b57d9e52a0e489153a0878",
    "id": "e6de71b25dbe89183bc2167e1396",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.507116317749023,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.044916782528162,
    "fitness": 0.8188733087433819,
    "legacy_fitness": 0.8188733087433819,
    "target_cdf": 0.9796648197280255,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.337939739227295,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7290450087932406,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.337939739227295,
        "normalized": 0.9796648197280255,
        "contribution": 0.4898324098640128
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5920509099960327,
        "normalized": 0.10422977602228944,
        "contribution": 0.02605744400557236
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.130748271942139,
        "normalized": 0.9142167170262676,
        "contribution": 0.13713250755394013
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.233165979385376,
        "normalized": 0.760226473697154,
        "contribution": 0.07602264736971541
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "da50408b1eecd94ccfe332f18d9f9cf168622cdbb87982051c0e3464650c1f99",
    "id": "e75d8e647cd42638ce8c96bb8195",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 3.8420867919921875,
    "wer": 0.0,
    "identity": 0.7687634825706482,
    "voiceclap": 0.13569135963916779,
    "fitness": 0.779079620366826,
    "legacy_fitness": 0.779079620366826,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7242896868147519,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0534186363220215,
        "normalized": 0.2337470068421033,
        "contribution": 0.05843675171052583
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.792710781097412,
        "normalized": 0.8577606879337987,
        "contribution": 0.1286641031900698
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.272770404815674,
        "normalized": 0.7710130971983604,
        "contribution": 0.07710130971983604
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9e540a01c90b703deb414843662626db76555e8e139d679fd1e91feac2d3d329",
    "id": "e7d4fc88b5fcce27dd4b145891dd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.55477237701416,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09325850009918213,
    "fitness": 0.8002106899447156,
    "legacy_fitness": 0.8002106899447156,
    "target_cdf": 0.946232810502631,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8834309577941895,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6858766204193514,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8834309577941895,
        "normalized": 0.946232810502631,
        "contribution": 0.4731164052513155
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.510633111000061,
        "normalized": 0.0882576514733458,
        "contribution": 0.02206441286833645
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.774919033050537,
        "normalized": 0.8542158295589685,
        "contribution": 0.12813237443384526
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.797515869140625,
        "normalized": 0.6256342786585414,
        "contribution": 0.06256342786585414
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9c84426c3d7ca9afcab6de76a83aaef339475350d355c56e3cbcae1b3b4aa32f",
    "id": "e915b2d147c89c1dc44242bb46a9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 4.419482707977295,
    "wer": 0.03333333333333333,
    "identity": 0.4135938882827759,
    "voiceclap": -0.024719802662730217,
    "fitness": 0.7825590125729013,
    "legacy_fitness": 0.7825590125729013,
    "target_cdf": 0.9445039705317666,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8671984672546387,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7176509846643164,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8671984672546387,
        "normalized": 0.9445039705317666,
        "contribution": 0.4722519852658833
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.161975860595703,
        "normalized": 0.27373275044956896,
        "contribution": 0.06843318761239224
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.321237087249756,
        "normalized": 0.744168371280294,
        "contribution": 0.1116252556920441
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8811254501342773,
        "normalized": 0.6534055609399674,
        "contribution": 0.06534055609399674
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "438af2b4ec6a27e2bc0038c37a01c14d575a123972b38f0b39f970b51e42b22d",
    "id": "e9578f8a70b1b8ae4cbaf3ac9cc6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.09409761428833,
    "genuineness": 4.356721878051758,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.016855264082551003,
    "fitness": 0.8110057119844014,
    "legacy_fitness": 0.8110057119844014,
    "target_cdf": 0.9542463867500731,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.9645934104919434,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7108601858469186,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.9645934104919434,
        "normalized": 0.9542463867500731,
        "contribution": 0.47712319337503656
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0602035522460938,
        "normalized": 0.23614752264221534,
        "contribution": 0.059036880660553834
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.476912260055542,
        "normalized": 0.786101975965994,
        "contribution": 0.11791529639489909
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6291964054107666,
        "normalized": 0.5678481541642914,
        "contribution": 0.05678481541642914
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4cf7daa96022418cdcaa8756e7828eef6d57f9cb969484c8ef4b8357b6016f21",
    "id": "e982438a00086dd3e1856e946607",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 4.582659721374512,
    "wer": 0.03571428571428571,
    "identity": 0.501045823097229,
    "voiceclap": -0.11200378090143204,
    "fitness": 0.7785333009888917,
    "legacy_fitness": 0.7785333009888917,
    "target_cdf": 0.9154342620847058,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.639944076538086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6443395965539812,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.639944076538086,
        "normalized": 0.9154342620847058,
        "contribution": 0.4577171310423529
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7548866271972656,
        "normalized": 0.14223933958928087,
        "contribution": 0.03555983489732022
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0654847621917725,
        "normalized": 0.6669986610607342,
        "contribution": 0.10004979915911014
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.465277671813965,
        "normalized": 0.5101283145519799,
        "contribution": 0.051012831455198
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b81d813c1f6693fc28417fc1143dbfa2305ed98603292f8c4f3616334e78a3bf",
    "id": "e9980fe169db7c5126c8501db957",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 3.6914620399475098,
    "wer": 0.0,
    "identity": 0.716664731502533,
    "voiceclap": 0.09889943897724152,
    "fitness": 0.7847173105297656,
    "legacy_fitness": 0.7847173105297656,
    "target_cdf": 0.9495623609658421,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.915895938873291,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7492062564877573,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.915895938873291,
        "normalized": 0.9495623609658421,
        "contribution": 0.47478118048292106
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.10091233253479,
        "normalized": 0.2508316293237907,
        "contribution": 0.06270790733094768
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8193979263305664,
        "normalized": 0.8629676950479268,
        "contribution": 0.12944515425718903
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4795942306518555,
        "normalized": 0.8227201441669949,
        "contribution": 0.0822720144166995
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "03b512544b85a1423df7ec3a44b4e090fbefef49b2961abb7186740735ee7fda",
    "id": "e99fef3692b420618e854a90c143",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2,
    "generation": null,
    "ce": 6.523001670837402,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03452805429697037,
    "fitness": 0.8164731728568573,
    "legacy_fitness": 0.8164731728568573,
    "target_cdf": 0.962564179804901,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.06198787689209,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6875087878623702,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.06198787689209,
        "normalized": 0.962564179804901,
        "contribution": 0.4812820899024505
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.524202823638916,
        "normalized": 0.09078486014132758,
        "contribution": 0.022696215035331896
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7571277618408203,
        "normalized": 0.8506121673397089,
        "contribution": 0.12759182510095632
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6049938201904297,
        "normalized": 0.5593865782363151,
        "contribution": 0.05593865782363151
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "6407338aa009066b00ee4e3a075782f01c8195e73a0a2f8a7d742552bcc2f386",
    "id": "e9dfed171d78463c02243eb04cd4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 3,
    "generation": null,
    "ce": 6.435632705688477,
    "genuineness": 3.5439748764038086,
    "wer": 0.0,
    "identity": 0.5458073019981384,
    "voiceclap": 0.09556063264608383,
    "fitness": 0.791673737917784,
    "legacy_fitness": 0.791673737917784,
    "target_cdf": 0.9662346008508267,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.110685348510742,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8265819778278539,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.110685348510742,
        "normalized": 0.9662346008508267,
        "contribution": 0.48311730042541334
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5962040424346924,
        "normalized": 0.4595519396623047,
        "contribution": 0.11488798491557617
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.07737398147583,
        "normalized": 0.9066346475271524,
        "contribution": 0.13599519712907285
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.06486177444458,
        "normalized": 0.9258149535779157,
        "contribution": 0.09258149535779157
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "874d490b8170a2dc3c4b2f66812f1e067559ab499e70fd02ae408d9d7535cc9d",
    "id": "eae7b2a9c6888bd32ed441fde209",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.287685871124268,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07564528286457062,
    "fitness": 0.7990362051383002,
    "legacy_fitness": 0.7990362051383002,
    "target_cdf": 0.9201750443886405,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6724090576171875,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6875027187133164,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6724090576171875,
        "normalized": 0.9201750443886405,
        "contribution": 0.46008752219432025
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8091652393341064,
        "normalized": 0.1567722644593298,
        "contribution": 0.03919306611483245
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4858078956604004,
        "normalized": 0.7883708195307028,
        "contribution": 0.11825562292960541
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0263421535491943,
        "normalized": 0.6996650747455826,
        "contribution": 0.06996650747455826
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "25128f0d1f8be00b3d23db9d39dfd55b6915347e251bc6fbf998047407482ba1",
    "id": "eba77b5e62b4a75931c63d215618",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.068024158477783,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0026663318276405334,
    "fitness": 0.8106810795798798,
    "legacy_fitness": 0.8106810795798798,
    "target_cdf": 0.962564179804901,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.06198787689209,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7185635676923848,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.06198787689209,
        "normalized": 0.962564179804901,
        "contribution": 0.4812820899024505
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.8638763427734375,
        "normalized": 0.8713534713164917,
        "contribution": 0.13070302069747375
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.175959348678589,
        "normalized": 0.7441683469516106,
        "contribution": 0.07441683469516107
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "54f3812b8aae14c806c30854b80594e7d6d53beadffd84db117fe1c7f0def24e",
    "id": "ec37fc8bbe5a739554beea684095",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 3,
    "generation": null,
    "ce": 6.348263263702393,
    "genuineness": 3.6789097785949707,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.13516438007354736,
    "fitness": 0.8098282640312172,
    "legacy_fitness": 0.8098282640312172,
    "target_cdf": 0.9956675499944005,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.938540458679199,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8496637369635227,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.938540458679199,
        "normalized": 0.9956675499944005,
        "contribution": 0.4978337749972003
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.53344464302063,
        "normalized": 0.431000315198613,
        "contribution": 0.10775007879965325
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.047008514404297,
        "normalized": 0.9850674568917444,
        "contribution": 0.14776011853376164
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.452106475830078,
        "normalized": 0.963197646329075,
        "contribution": 0.0963197646329075
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4769ba85e10b31c6a9b06da0b20aa927abb3bbe9e0ff47206c180418bebef6c4",
    "id": "ec8dc745c1c2490ca5dcdb7625df",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": 0.21042412519454956,
    "voiceclap": -0.06858615577220917,
    "fitness": 0.7715818205727856,
    "legacy_fitness": 0.7715818205727856,
    "target_cdf": 0.8542158920076929,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.3071789741516113,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6905766893067948,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.3071789741516113,
        "normalized": 0.8542158920076929,
        "contribution": 0.42710794600384644
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.3400771617889404,
        "normalized": 0.34587384429467316,
        "contribution": 0.08646846107366829
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.272310495376587,
        "normalized": 0.730154794161955,
        "contribution": 0.10952321912429323
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9471330642700195,
        "normalized": 0.674770631049869,
        "contribution": 0.0674770631049869
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8711347a858576190bfcf05e60e20c0ef1ed26cdd95a2a36abe2c2ef8b038210",
    "id": "ecdbfdb677d5c2f520d4fa6f5bd1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 4.218649387359619,
    "wer": 0.0,
    "identity": 0.6665695905685425,
    "voiceclap": 0.08335374295711517,
    "fitness": 0.8324653718557468,
    "legacy_fitness": 0.8324653718557468,
    "target_cdf": 0.9995376693579576,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.652768611907959,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7789243558364097,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.652768611907959,
        "normalized": 0.9995376693579576,
        "contribution": 0.4997688346789788
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.856658935546875,
        "normalized": 0.17026338423128196,
        "contribution": 0.04256584605782049
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.7979278564453125,
        "normalized": 0.974597599214758,
        "contribution": 0.1461896398822137
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.9064433574676514,
        "normalized": 0.9040003521739675,
        "contribution": 0.09040003521739676
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "625073b4f221ad711c5104fd21353dda28da53115ac6f8f010e078df2e85ed3f",
    "id": "ed59c567082045344c3d63216c68",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.07972566783428192,
    "fitness": 0.832574138464476,
    "legacy_fitness": 0.832574138464476,
    "target_cdf": 0.98254389250665,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.402869701385498,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7601583733994317,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.402869701385498,
        "normalized": 0.98254389250665,
        "contribution": 0.491271946253325
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0839502811431885,
        "normalized": 0.24465512789027288,
        "contribution": 0.06116378197256822
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7215447425842285,
        "normalized": 0.8432277384318043,
        "contribution": 0.12648416076477065
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.43558931350708,
        "normalized": 0.8123848440876791,
        "contribution": 0.08123848440876791
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f98d328bb9e26e0bede09bb23a59c6084f89fc72d541a97863f3f5a04c99c86a",
    "id": "ed889447dab5e57a6d520fd46b38",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.025177273899316788,
    "fitness": 0.8293297493856995,
    "legacy_fitness": 0.8293297493856995,
    "target_cdf": 0.9896249303026692,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6138916015625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7292118473200363,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6138916015625,
        "normalized": 0.9896249303026692,
        "contribution": 0.4948124651513346
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7141776084899902,
        "normalized": 0.13195706015785413,
        "contribution": 0.03298926503946353
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.828293561935425,
        "normalized": 0.8646740370257361,
        "contribution": 0.1297011055538604
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0835487842559814,
        "normalized": 0.7170901157537775,
        "contribution": 0.07170901157537775
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "33c6ffc2b3a6f5207d5abfae2fc76d93afdc028c0a946715f7adb814e7a09430",
    "id": "ee06bba7871406a7f2d796ba8fee",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 3,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": 0.7375544309616089,
    "voiceclap": 0.06292987614870071,
    "fitness": 0.8105941425429051,
    "legacy_fitness": 0.8105941425429051,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7889300822485775,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1517984867095947,
        "normalized": 0.26984519049493305,
        "contribution": 0.06746129762373326
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.495473384857178,
        "normalized": 0.9542463876094935,
        "contribution": 0.14313695814142402
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8360352516174316,
        "normalized": 0.8929162948189704,
        "contribution": 0.08929162948189705
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5e95436049cb75e064621d68f4464aa1ce2cf000a9dd1a6722379b7801f52835",
    "id": "eeff8bbc9bd61a86cfcc7e21f236",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 4.250029563903809,
    "wer": 0.0,
    "identity": 0.6595120429992676,
    "voiceclap": -0.0740094929933548,
    "fitness": 0.8233919291531215,
    "legacy_fitness": 0.8233919291531215,
    "target_cdf": 0.9763929662774343,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.273009777069092,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7196058513502651,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.273009777069092,
        "normalized": 0.9763929662774343,
        "contribution": 0.48819648313871716
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8227348327636719,
        "normalized": 0.16055304509332224,
        "contribution": 0.04013826127333056
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4858078956604004,
        "normalized": 0.7883708195307028,
        "contribution": 0.11825562292960541
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.127553939819336,
        "normalized": 0.7301548400861202,
        "contribution": 0.07301548400861202
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "834686513d280e085219a3c3e1f2d1305793546f92d7066e69bc94fd38db868c",
    "id": "f0a8f695ff46ab59abb6a36b5d7d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.0003551645204424858,
    "fitness": 0.8288818059488978,
    "legacy_fitness": 0.8288818059488978,
    "target_cdf": 0.994317161664336,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.8411459922790527,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7612272632927715,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.8411459922790527,
        "normalized": 0.994317161664336,
        "contribution": 0.497158580832168
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8227348327636719,
        "normalized": 0.16055304509332224,
        "contribution": 0.04013826127333056
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.335350036621094,
        "normalized": 0.93905329606414,
        "contribution": 0.140857994409621
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.514798402786255,
        "normalized": 0.8307242677765183,
        "contribution": 0.08307242677765184
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "8b35044a7a1ec480bc0b2b1f9839d4009842b2a600274b1978371de77cb4fdfd",
    "id": "f11e11efaf4fd28ece22064e5913",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.133810997009277,
    "genuineness": 4.093128204345703,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.02439941093325615,
    "fitness": 0.8045275807050842,
    "legacy_fitness": 0.8045275807050842,
    "target_cdf": 0.961270594530896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.045755386352539,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8477686329747425,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.045755386352539,
        "normalized": 0.961270594530896,
        "contribution": 0.480635297265448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.7912676334381104,
        "normalized": 0.5489650027609123,
        "contribution": 0.13724125069022808
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "024f35a8f6b5c07a660cbc3f77e6775c79c05bae955b231e75751a000a846ed8",
    "id": "f1504463c25efd47ca49217e60f5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.260893821716309,
    "genuineness": 4.168440818786621,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.01583227701485157,
    "fitness": 0.829962962116439,
    "legacy_fitness": 0.829962962116439,
    "target_cdf": 0.9995111649148442,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.636536121368408,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7557638033256446,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.636536121368408,
        "normalized": 0.9995111649148442,
        "contribution": 0.4997555824574221
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5309876203536987,
        "normalized": 0.09206841268091792,
        "contribution": 0.02301710317022948
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.842406749725342,
        "normalized": 0.9768247598751261,
        "contribution": 0.1465237139812689
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.677617073059082,
        "normalized": 0.8646740371672423,
        "contribution": 0.08646740371672423
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "546ef80033171d090672724af75e3641b47415ac90d2d5e630ac16b281812a97",
    "id": "f19c8a7c7e52c9b46334dad0e811",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.205295085906982,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": 0.45037615299224854,
    "voiceclap": -0.17895819246768951,
    "fitness": 0.8110684809912161,
    "legacy_fitness": 0.8110684809912161,
    "target_cdf": 0.9519500370480449,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.940244674682617,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6941874499570295,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.940244674682617,
        "normalized": 0.9519500370480449,
        "contribution": 0.47597501852402246
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7413170337677002,
        "normalized": 0.13875333808010226,
        "contribution": 0.034688334520025565
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.34792423248291,
        "normalized": 0.7516488247877673,
        "contribution": 0.1127473237181651
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.0527453422546387,
        "normalized": 0.7077677319481637,
        "contribution": 0.07077677319481637
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a80df25a1c4f827cdea8a6a94965f011af1b14aea6df13220cb1324bbc7da476",
    "id": "f1bd1d697e058e720f5b6c804728",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.419482707977295,
    "wer": 0.0,
    "identity": 0.39370474219322205,
    "voiceclap": -0.09421361982822418,
    "fitness": 0.8386085494843268,
    "legacy_fitness": 0.8386085494843268,
    "target_cdf": 0.9867481280341037,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5164966583251953,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.761520017549604,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5164966583251953,
        "normalized": 0.9867481280341037,
        "contribution": 0.49337406401705186
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.009317398071289,
        "normalized": 0.21847802587457404,
        "contribution": 0.05461950646864351
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.015103816986084,
        "normalized": 0.8971763038367273,
        "contribution": 0.1345764455755091
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3431787490844727,
        "normalized": 0.7895000148839948,
        "contribution": 0.07895000148839948
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "67608dec8ab0d9f76e1fd0d9c513424488d6110fe03a81468665367be8b46759",
    "id": "f1dd2bcd8c4e64dda34849e097fd",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.024319220334291458,
    "fitness": 0.818955616826273,
    "legacy_fitness": 0.818955616826273,
    "target_cdf": 0.9696036431679269,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1593828201293945,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7628274405688287,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1593828201293945,
        "normalized": 0.9696036431679269,
        "contribution": 0.48480182158396345
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1517984867095947,
        "normalized": 0.26984519049493305,
        "contribution": 0.06746129762373326
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.050686836242676,
        "normalized": 0.9026629058044647,
        "contribution": 0.13539943587066972
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.202362537384033,
        "normalized": 0.7516488549046221,
        "contribution": 0.07516488549046221
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0599c1d5191e9f543e19f804dd94338e2d251fced2cc7bf7ba8e8e98c0c2a1b0",
    "id": "f1dd98b65dcf7e0ab51405cfe968",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.13078498840332,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.037831708788871765,
    "fitness": 0.8159267466777058,
    "legacy_fitness": 0.8159267466777058,
    "target_cdf": 0.9650455080496254,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.0944528579711914,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7570105357466188,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.0944528579711914,
        "normalized": 0.9650455080496254,
        "contribution": 0.4825227540248127
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1212668418884277,
        "normalized": 0.25835102183826425,
        "contribution": 0.06458775545956606
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.006208419799805,
        "normalized": 0.8957702550325092,
        "contribution": 0.13436553825487638
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.215564012527466,
        "normalized": 0.7553448800736362,
        "contribution": 0.07553448800736362
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "68d9ca3f3d753b6d092d520978e36af2da272545c672aaca596ba269a6c23c2a",
    "id": "f20ffae6b3eede46e4485069c0ca",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 3.7918784618377686,
    "wer": 0.0,
    "identity": 0.7153233289718628,
    "voiceclap": 0.04829482361674309,
    "fitness": 0.8039708145201723,
    "legacy_fitness": 0.8039708145201723,
    "target_cdf": 0.9673902089874984,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.126917839050293,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7383037580356065,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.126917839050293,
        "normalized": 0.9673902089874984,
        "contribution": 0.4836951044937492
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.121852874755859,
        "normalized": 0.9129861397194918,
        "contribution": 0.13694792095792377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.360780715942383,
        "normalized": 0.7939810260608104,
        "contribution": 0.07939810260608104
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ea0a65e5b0843fc5b731135be7fe1344489c51aebd142cf44c78cc73389dc3f5",
    "id": "f2c4051087f6fb7b624bead4548a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1,
    "generation": null,
    "ce": 6.260893821716309,
    "genuineness": 4.193545341491699,
    "wer": 0.03571428571428571,
    "identity": null,
    "voiceclap": -0.06416326016187668,
    "fitness": 0.7999860605790917,
    "legacy_fitness": 0.7999860605790917,
    "target_cdf": 0.9967235449909552,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.035935401916504,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7624000806884861,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.035935401916504,
        "normalized": 0.9967235449909552,
        "contribution": 0.4983617724954776
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7006080150604248,
        "normalized": 0.12864648958919778,
        "contribution": 0.032161622397299444
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.48657751083374,
        "normalized": 0.9534909337644679,
        "contribution": 0.1430236400646702
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8096323013305664,
        "normalized": 0.8885304573103883,
        "contribution": 0.08885304573103883
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e296acdc40f7a82a85fe270cb63370f474781427368c2e83cbd0d60a2b7a267e",
    "id": "f3630f3ed42dbca0bb30616eeaf6",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1,
    "generation": null,
    "ce": 6.4912309646606445,
    "genuineness": 4.369274139404297,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.10923629254102707,
    "fitness": 0.8174502685096773,
    "legacy_fitness": 0.8174502685096773,
    "target_cdf": 0.946232810502631,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.8834309577941895,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6944220731278817,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.8834309577941895,
        "normalized": 0.946232810502631,
        "contribution": 0.4731164052513155
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.707392930984497,
        "normalized": 0.130294529860791,
        "contribution": 0.03257363246519775
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5436301231384277,
        "normalized": 0.8027714861707109,
        "contribution": 0.12041572292560664
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9735360145568848,
        "normalized": 0.6831631248576189,
        "contribution": 0.0683163124857619
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "c30b314a2029b8aca8e9897a2d061f9fea64aaa3a6e06f4a3b34bff40e51f2ff",
    "id": "f36399553115e1997561dd2ae6f0",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.09409761428833,
    "genuineness": 4.331617832183838,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.04240422323346138,
    "fitness": 0.8256600028866828,
    "legacy_fitness": 0.8256600028866828,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7664321376424386,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9889628887176514,
        "normalized": 0.21162916044311025,
        "contribution": 0.05290729011077756
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.121852874755859,
        "normalized": 0.9129861397194918,
        "contribution": 0.13694792095792377
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5456018447875977,
        "normalized": 0.8375344038525582,
        "contribution": 0.08375344038525583
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bad01b9dc9f0b4024f27b0b2a9a30861f8a4f353bfc9afc3eb4054d73cdb7d09",
    "id": "f37f92ab45bb9e65b985c149f9b5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.125868320465088,
    "genuineness": 4.086852550506592,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.1403050720691681,
    "fitness": 0.8232854531645789,
    "legacy_fitness": 0.8232854531645789,
    "target_cdf": 0.9997064444303541,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.782628059387207,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.800537801052526,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.782628059387207,
        "normalized": 0.9997064444303541,
        "contribution": 0.49985322221517703
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0873427391052246,
        "normalized": 0.2458838571357379,
        "contribution": 0.061470964283934476
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.029216766357422,
        "normalized": 0.9844679187583991,
        "contribution": 0.14767018781375987
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.985652446746826,
        "normalized": 0.9154342673965461,
        "contribution": 0.09154342673965461
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0ff6005d6cf95a9bc9509475f333b9d65dbfb99bcfe11c17c8bee642a0841ce0",
    "id": "f41cbbd8785378b2a874d3b98a49",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.306513786315918,
    "wer": 0.0,
    "identity": 0.7550034523010254,
    "voiceclap": -0.03844553232192993,
    "fitness": 0.8321736383926128,
    "legacy_fitness": 0.8321736383926128,
    "target_cdf": 0.9932207044941694,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7762160301208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7756362619376704,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7762160301208496,
        "normalized": 0.9932207044941694,
        "contribution": 0.4966103522470847
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.961823582649231,
        "normalized": 0.2026956548789064,
        "contribution": 0.0506739137197266
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.317558765411377,
        "normalized": 0.9371462024575616,
        "contribution": 0.14057193036863425
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7480251789093018,
        "normalized": 0.877800656022249,
        "contribution": 0.0877800656022249
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b2cd9d14534e42362a720bda81108e4f2aeae76b82e56974714556886a804952",
    "id": "f5568254212d8732e4cdcb09fb84",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.281409740447998,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.1893063634634018,
    "fitness": 0.7010803350051258,
    "legacy_fitness": 0.7010803350051258,
    "target_cdf": 0.7275662279605893,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 1.8404958248138428,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6063264520648713,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 1.8404958248138428,
        "normalized": 0.7275662279605893,
        "contribution": 0.36378311398029467
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1721529960632324,
        "normalized": 0.2776477208574187,
        "contribution": 0.06941193021435467
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.0588130950927734,
        "normalized": 0.6648663906981253,
        "contribution": 0.09972995860471878
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.1407554149627686,
        "normalized": 0.7340144926550323,
        "contribution": 0.07340144926550322
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "950c08ebd8e0b9dd0abc25f1c4169fc9beee9ca0d783c1cc6fb118230c1496cb",
    "id": "f55c87a2e9e345c22fcdf863a5d9",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 4.469690799713135,
    "wer": 0.0,
    "identity": 0.3901166319847107,
    "voiceclap": 0.014104008674621582,
    "fitness": 0.8255676231917586,
    "legacy_fitness": 0.8255676231917586,
    "target_cdf": 0.9592621807529222,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.021406888961792,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7019935626045495,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.021406888961792,
        "normalized": 0.9592621807529222,
        "contribution": 0.4796310903764611
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8295196294784546,
        "normalized": 0.16246558721860938,
        "contribution": 0.040616396804652344
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4057464599609375,
        "normalized": 0.767448144534375,
        "contribution": 0.11511722168015626
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9207301139831543,
        "normalized": 0.6662885374327974,
        "contribution": 0.06662885374327975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "93b83d85bb0c2b13ed2320bc280d782cff28d828e09d9cb448b208dd4819c2a9",
    "id": "f57396ec93f20949f9f7486c4047",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 2,
    "generation": null,
    "ce": 6.086154937744141,
    "genuineness": 4.362998008728027,
    "wer": 0.06666666666666667,
    "identity": 0.22995468974113464,
    "voiceclap": -0.14788004755973816,
    "fitness": 0.7213227240898477,
    "legacy_fitness": 0.7213227240898477,
    "target_cdf": 0.8778006373398928,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.4208059310913086,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6489309131704457,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.4208059310913086,
        "normalized": 0.8778006373398928,
        "contribution": 0.4389003186699464
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9245071411132812,
        "normalized": 0.1907869738584032,
        "contribution": 0.0476967434646008
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.087724208831787,
        "normalized": 0.6740671752112604,
        "contribution": 0.10111007628168905
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.757911205291748,
        "normalized": 0.6122377475420941,
        "contribution": 0.06122377475420942
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "370078ab3f8a3271c02601e732fe7d94f87754c4e3fa57c574c9ba4964b7a685",
    "id": "f57baddc93d8a619604ab295bade",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.3641486167907715,
    "genuineness": 4.268857479095459,
    "wer": 0.0,
    "identity": 0.7139122486114502,
    "voiceclap": -0.1296544075012207,
    "fitness": 0.8285761271158649,
    "legacy_fitness": 0.8285761271158649,
    "target_cdf": 0.9832067001342365,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.419102191925049,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7359622326557521,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.419102191925049,
        "normalized": 0.9832067001342365,
        "contribution": 0.49160335006711825
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8396968841552734,
        "normalized": 0.16536210020893177,
        "contribution": 0.04134052505223294
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.7037534713745117,
        "normalized": 0.8394469854437661,
        "contribution": 0.1259170478165649
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.272770404815674,
        "normalized": 0.7710130971983604,
        "contribution": 0.07710130971983604
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4542026fd649cb46ef4db4abbb5a817cd965321f6e49f58869033d42a7e1bc8d",
    "id": "f592924a2fe322cefba338bb8396",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.07942438125610352,
    "fitness": 0.8249166033728828,
    "legacy_fitness": 0.8249166033728828,
    "target_cdf": 0.9968750600147067,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.052167892456055,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7685169117905717,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.052167892456055,
        "normalized": 0.9968750600147067,
        "contribution": 0.49843753000735336
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8634437322616577,
        "normalized": 0.17224968684334863,
        "contribution": 0.043062421710837157
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.646700859069824,
        "normalized": 0.9656443012467251,
        "contribution": 0.14484664518700877
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.475193977355957,
        "normalized": 0.8217031488537243,
        "contribution": 0.08217031488537244
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fcab01004cce44ea505cc621b620c6cf8d01910b2e86e6fc3e3e1aa53f042c21",
    "id": "f5b6b3c9929970bf970254f975ab",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 3,
    "generation": null,
    "ce": 6.078212261199951,
    "genuineness": 3.7165660858154297,
    "wer": 0.0,
    "identity": 0.5852750539779663,
    "voiceclap": 0.13750103116035461,
    "fitness": 0.8068000892984356,
    "legacy_fitness": 0.8068000892984356,
    "target_cdf": 0.9999757250522545,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 5.431926727294922,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8340904841840501,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 5.431926727294922,
        "normalized": 0.9999757250522545,
        "contribution": 0.49998786252612726
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.334988594055176,
        "normalized": 0.3437155282365555,
        "contribution": 0.08592888205913887
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.705292701721191,
        "normalized": 0.9970202396210475,
        "contribution": 0.1495530359431571
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.918560028076172,
        "normalized": 0.9862070365562678,
        "contribution": 0.09862070365562679
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "9e42f56ea075384307c690a4b92a4654b870879a66f180748558bee141b4734c",
    "id": "f6c4860d243e96c699cfc4f725ca",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2,
    "generation": null,
    "ce": 6.594485759735107,
    "genuineness": 4.319066047668457,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.09602750837802887,
    "fitness": 0.836426579380015,
    "legacy_fitness": 0.836426579380015,
    "target_cdf": 0.9832067001342365,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.419102191925049,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7696919831783852,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.419102191925049,
        "normalized": 0.9832067001342365,
        "contribution": 0.49160335006711825
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.066988229751587,
        "normalized": 0.2385614708758641,
        "contribution": 0.059640367718966025
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.139644145965576,
        "normalized": 0.9154342857630313,
        "contribution": 0.1373151428644547
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4311888217926025,
        "normalized": 0.8113312252784615,
        "contribution": 0.08113312252784616
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3cbd9fb30ceb85ae1871f915a051b9ac68496cc0b3eba82b07cfb583293e6fa9",
    "id": "f6dd3294ebbd43f60bbd8d6c2689",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Du bist der erste Mensch, den ich heute Abend finden wollte. Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig. Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "sentences": [
      "Du bist der erste Mensch, den ich heute Abend finden wollte.",
      "Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.",
      "Bitte hör genau zu, denn das ist wichtiger, als ich dachte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Du bist der erste Mensch, den ich heute Abend finden wollte.\n(Natural clear speech.) [4.7 seconds duration] Ich wollte mich darauf vorbereiten... aber die Worte verändern sich ständig.\n(Natural clear speech.) [4.7 seconds duration] Bitte hör genau zu, denn das ist wichtiger, als ich dachte.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 3.9676074981689453,
    "wer": 0.0,
    "identity": 0.6747179627418518,
    "voiceclap": -0.00038238801062107086,
    "fitness": 0.8234254643718518,
    "legacy_fitness": 0.8234254643718518,
    "target_cdf": 0.999200941205048,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.490444183349609,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.774792232329725,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.490444183349609,
        "normalized": 0.999200941205048,
        "contribution": 0.499600470602524
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8532664775848389,
        "normalized": 0.16927574075204688,
        "contribution": 0.04231893518801172
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.6555962562561035,
        "normalized": 0.9662345967558206,
        "contribution": 0.14493518951337309
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.756826162338257,
        "normalized": 0.8793763702581625,
        "contribution": 0.08793763702581625
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "fcaefbcad65a8a7407e2269458cd6b100681f6cfbcdb4292405c57f0d7a61f3c",
    "id": "f6dde6cde763581b38d17c955a56",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.193545341491699,
    "wer": 0.0,
    "identity": 0.6668914556503296,
    "voiceclap": 0.036773186177015305,
    "fitness": 0.7822785872609868,
    "legacy_fitness": 0.7822785872609868,
    "target_cdf": 0.8929163174612232,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5019683837890625,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6708058424747226,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5019683837890625,
        "normalized": 0.8929163174612232,
        "contribution": 0.4464581587306116
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8939753770828247,
        "normalized": 0.18137005253541438,
        "contribution": 0.045342513133853596
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.3123412132263184,
        "normalized": 0.7416489601774914,
        "contribution": 0.1112473440266237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9559340476989746,
        "normalized": 0.6775782658363374,
        "contribution": 0.06775782658363375
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "468e16e82f3d40b2299444c8320766dec23bfb393f52477a870f5e9f781086e7",
    "id": "f7b0c68a5eb61c7801b83b10bbcb",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": true,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.570107460021973,
    "wer": 0.0,
    "identity": 0.20552334189414978,
    "voiceclap": -0.17014914751052856,
    "fitness": 0.8134540133310864,
    "legacy_fitness": 0.8134540133310864,
    "target_cdf": 0.929049708969098,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.7373390197753906,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6707200637911697,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.7373390197753906,
        "normalized": 0.929049708969098,
        "contribution": 0.464524854484549
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.734532117843628,
        "normalized": 0.13703231036620456,
        "contribution": 0.03425807759155114
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4190900325775146,
        "normalized": 0.7710131225645452,
        "contribution": 0.11565196838468178
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.6148948669433594,
        "normalized": 0.562851633303878,
        "contribution": 0.0562851633303878
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d00a0353f8b18a760a8937c05563e3dc783b45ac91c92964afba5fe327303897",
    "id": "f7c00364ac7df56c3ca5f5d32ce4",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.475345611572266,
    "genuineness": 4.494794845581055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.18211150169372559,
    "fitness": 0.8359957018603179,
    "legacy_fitness": 0.8359957018603179,
    "target_cdf": 0.9736578860102679,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.2243127822875977,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7344415388469326,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.2243127822875977,
        "normalized": 0.9736578860102679,
        "contribution": 0.48682894300513396
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8770133256912231,
        "normalized": 0.17626644875656883,
        "contribution": 0.04406661218914221
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.6770663261413574,
        "normalized": 0.8336650598974741,
        "contribution": 0.12504975898462112
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3255765438079834,
        "normalized": 0.7849622466803536,
        "contribution": 0.07849622466803537
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "f7782a460b3b2d322a4e0de695ac02113aa194a471f16eb749122d8941430b26",
    "id": "f817a2f7c40153966dc181401d9f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.300607204437256,
    "genuineness": 4.243753433227539,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0765630453824997,
    "fitness": 0.8282260030459647,
    "legacy_fitness": 0.8282260030459647,
    "target_cdf": 0.9877755264344382,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.548961639404297,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7616447237303919,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.548961639404297,
        "normalized": 0.9877755264344382,
        "contribution": 0.4938877632172191
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9041526317596436,
        "normalized": 0.18447617999078308,
        "contribution": 0.04611904499769577
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.139644145965576,
        "normalized": 0.9154342857630313,
        "contribution": 0.1373151428644547
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.572004795074463,
        "normalized": 0.8432277265102239,
        "contribution": 0.08432277265102239
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "ba9c2b3c501c36007f1d1739ef66944c7c21bd81c4155052eb2a4ea46e4d15ed",
    "id": "f81caf506068af1be5aea35e8d6b",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": false,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 2,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.9676074981689453,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.027908841148018837,
    "fitness": 0.8172154341379597,
    "legacy_fitness": 0.8172154341379597,
    "target_cdf": 0.9832067001342365,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.419102191925049,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7391992475490314,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.419102191925049,
        "normalized": 0.9832067001342365,
        "contribution": 0.49160335006711825
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.673468828201294,
        "normalized": 0.12219935202142257,
        "contribution": 0.030549838005355642
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.246392726898193,
        "normalized": 0.9290496824330181,
        "contribution": 0.1393574523649527
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2947731018066406,
        "normalized": 0.7768860711160478,
        "contribution": 0.07768860711160479
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1d2681cd7ef47eef3096391e0de33adc4cbf7cb8e8c38638fca96627424ebd42",
    "id": "f8523495371e7057e4ad6da1c25f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.356205940246582,
    "genuineness": 4.632867813110352,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0227012000977993,
    "fitness": 0.8039414854245281,
    "legacy_fitness": 0.8039414854245281,
    "target_cdf": 0.9040003560775312,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5668981075286865,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6822489169426996,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5668981075286865,
        "normalized": 0.9040003560775312,
        "contribution": 0.4520001780387656
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.063595771789551,
        "normalized": 0.23735276923124937,
        "contribution": 0.05933819230781234
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.2011446952819824,
        "normalized": 0.7091081777136116,
        "contribution": 0.10636622665704173
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8569226264953613,
        "normalized": 0.6454431993908001,
        "contribution": 0.06454431993908001
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4dceda0ccbe575e87e98283e0e88b255f64bd28b24c06f5639ca0afa524804de",
    "id": "f86f03c95943c32c2e46b9cca98c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.332377910614014,
    "genuineness": 4.457139015197754,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.06304287165403366,
    "fitness": 0.8406339256456747,
    "legacy_fitness": 0.8406339256456747,
    "target_cdf": 0.9932207044941694,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7762160301208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7888440850837731,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7762160301208496,
        "normalized": 0.9932207044941694,
        "contribution": 0.4966103522470847
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.10091233253479,
        "normalized": 0.2508316293237907,
        "contribution": 0.06270790733094768
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.450994491577148,
        "normalized": 0.9503685278749294,
        "contribution": 0.14255527918123942
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.7040200233459473,
        "normalized": 0.8697054632450133,
        "contribution": 0.08697054632450134
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "bf6f71e501132ae7d95d851cdff3594c335fdf65ef9ea817272e4b2873bff531",
    "id": "f874e6fdb79fd9e56ed0fb5fc2ec",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": 0.6173027157783508,
    "voiceclap": -0.04859335720539093,
    "fitness": 0.8251091718343873,
    "legacy_fitness": 0.8251091718343873,
    "target_cdf": 0.961270594530896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.045755386352539,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7282499252694092,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.045755386352539,
        "normalized": 0.961270594530896,
        "contribution": 0.480635297265448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.1178743839263916,
        "normalized": 0.25708971381132506,
        "contribution": 0.06427242845283127
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.427985668182373,
        "normalized": 0.7733726160960801,
        "contribution": 0.116005892414412
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.942732572555542,
        "normalized": 0.67336307136718,
        "contribution": 0.067336307136718
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "19f2afdef541b1522cea77b65b2e8143c629aac152bafc7d61229e99da29844c",
    "id": "f887828b09285b54e898846b0298",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.04,
    "cfg": 2,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.105680465698242,
    "wer": 0.0,
    "identity": 0.6116994023323059,
    "voiceclap": -0.048499610275030136,
    "fitness": 0.7803504261606095,
    "legacy_fitness": 0.7803504261606095,
    "target_cdf": 0.8971763358976756,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.5263171195983887,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.693836532194045,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.5263171195983887,
        "normalized": 0.8971763358976756,
        "contribution": 0.4485881679488378
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.948253870010376,
        "normalized": 0.19831469683683772,
        "contribution": 0.04957867420920943
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.4235379695892334,
        "normalized": 0.7721946200719969,
        "contribution": 0.11582919301079952
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.378382682800293,
        "normalized": 0.7984049702519815,
        "contribution": 0.07984049702519816
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "3b9b62c292b8e0fc68111bb75c35c352d70f2136c8449376cbd799d6f025e0d2",
    "id": "f921610530d75f8a87ab948bf0ec",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.308549880981445,
    "genuineness": 4.143336772918701,
    "wer": 0.0,
    "identity": 0.688468337059021,
    "voiceclap": -0.042437583208084106,
    "fitness": 0.8269631314765667,
    "legacy_fitness": 0.8269631314765667,
    "target_cdf": 0.9932207044941694,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.7762160301208496,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8216459565583897,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.7762160301208496,
        "normalized": 0.9932207044941694,
        "contribution": 0.4966103522470847
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.4435458183288574,
        "normalized": 0.3907581380976326,
        "contribution": 0.09768953452440815
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.184122562408447,
        "normalized": 0.9213282021786904,
        "contribution": 0.13819923032680356
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8272342681884766,
        "normalized": 0.891468394600932,
        "contribution": 0.0891468394600932
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "166fcad6a0b589df03699358bd7a4c755f743c58b70ae426477ea1eac1f3e0aa",
    "id": "f97e344e3a26c3da49b8c8ab0b5c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 2,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 3.8671910762786865,
    "wer": 0.0,
    "identity": 0.7540922164916992,
    "voiceclap": 0.005761142820119858,
    "fitness": 0.7816696084942363,
    "legacy_fitness": 0.7816696084942363,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6740052319867114,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5716965198516846,
        "normalized": 0.1000529371132764,
        "contribution": 0.0250132342783191
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.650379180908203,
        "normalized": 0.8277503599224988,
        "contribution": 0.12416255398837481
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.8283193111419678,
        "normalized": 0.6359507907076711,
        "contribution": 0.06359507907076711
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "97852bf8a4289a6e21c654e30300f277749659dcc1cb04ca734116b3f988a55d",
    "id": "f9b9ebf0718e878c856f1bfdf65f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "de",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.340320587158203,
    "genuineness": 4.482243061065674,
    "wer": 0.0,
    "identity": 0.6636070013046265,
    "voiceclap": 0.01653633452951908,
    "fitness": 0.8380916284118396,
    "legacy_fitness": 0.8380916284118396,
    "target_cdf": 0.9856469723769627,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.4840316772460938,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7453891156484254,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.4840316772460938,
        "normalized": 0.9856469723769627,
        "contribution": 0.4928234861884814
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.0161020755767822,
        "normalized": 0.22078898327927737,
        "contribution": 0.05519724581981934
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.854980707168579,
        "normalized": 0.8697054717721656,
        "contribution": 0.13045582076582485
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9295310974121094,
        "normalized": 0.6691256287429975,
        "contribution": 0.06691256287429975
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "78716957532ace4ab23faf3f22b83a969e0990fe5fd504b5be9186fdf6a3e423",
    "id": "f9cbb329cdbcfd91e5edcd2fb3fc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 3.8797430992126465,
    "wer": 0.0,
    "identity": 0.6291650533676147,
    "voiceclap": 0.004468478262424469,
    "fitness": 0.7959379921970935,
    "legacy_fitness": 0.7959379921970935,
    "target_cdf": 0.9495623609658421,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.915895938873291,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6912815878564186,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.915895938873291,
        "normalized": 0.9495623609658421,
        "contribution": 0.47478118048292106
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6259750127792358,
        "normalized": 0.11146953771947371,
        "contribution": 0.027867384429868428
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.699305534362793,
        "normalized": 0.8384925416157307,
        "contribution": 0.1257738812423596
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.80631685256958,
        "normalized": 0.6285914170126945,
        "contribution": 0.06285914170126945
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "a1f849736e41f566e60f3eac784b12852e08b29c9dc98a0d38baaaf119934830",
    "id": "fa147e3e22cd1b9a32c647377aab",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.4515180587768555,
    "genuineness": 4.293961524963379,
    "wer": 0.0,
    "identity": 0.5536198616027832,
    "voiceclap": -0.09900252521038055,
    "fitness": 0.8390883642151745,
    "legacy_fitness": 0.8390883642151745,
    "target_cdf": 0.9977706984112247,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.165794849395752,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7388417989199935,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.165794849395752,
        "normalized": 0.9977706984112247,
        "contribution": 0.49888534920561234
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5581269264221191,
        "normalized": 0.09733711108070414,
        "contribution": 0.024334277770176035
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.246392726898193,
        "normalized": 0.9290496824330181,
        "contribution": 0.1393574523649527
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.241966962814331,
        "normalized": 0.762647195792524,
        "contribution": 0.0762647195792524
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "832f9031d9b30199a5a20fedf699034364e9a58b8f15d90cb70af5811d926e78",
    "id": "fa1b13f1d5c2606703f554ab484f",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 3,
    "generation": null,
    "ce": 6.324435234069824,
    "genuineness": 3.9676074981689453,
    "wer": 0.0,
    "identity": 0.5091354846954346,
    "voiceclap": 0.08411435782909393,
    "fitness": 0.8233244597882228,
    "legacy_fitness": 0.8233244597882228,
    "target_cdf": 0.9997931996922089,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.880023002624512,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8581549313908451,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.880023002624512,
        "normalized": 0.9997931996922089,
        "contribution": 0.49989659984610446
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.5597357749938965,
        "normalized": 0.4429260890535954,
        "contribution": 0.11073152226339884
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 5.651917934417725,
        "normalized": 0.996565454227725,
        "contribution": 0.14948481813415876
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.760141372680664,
        "normalized": 0.9804199114718302,
        "contribution": 0.09804199114718302
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0567952a5afbd825d18b4f840ac34893ee439009893a08721313723d3963824c",
    "id": "fa838faf5dd42bb0210820db1023",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 4.017816066741943,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.0905744731426239,
    "fitness": 0.8140887610468241,
    "legacy_fitness": 0.8140887610468241,
    "target_cdf": 0.9711807712620442,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.1837315559387207,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7305425484353165,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.1837315559387207,
        "normalized": 0.9711807712620442,
        "contribution": 0.4855903856310221
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6870384216308594,
        "normalized": 0.12539399986239474,
        "contribution": 0.031348499965598686
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.068478107452393,
        "normalized": 0.9053242349151492,
        "contribution": 0.13579863523727237
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.299173593521118,
        "normalized": 0.7780502760142334,
        "contribution": 0.07780502760142334
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "e863cb0a69587f4277bb3931ec0b829674cc6188845a6232e4df614f0fd76b10",
    "id": "fb9744ce86c8788fbdcc935f3f23",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.372091293334961,
    "genuineness": 3.6914620399475098,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.09736888855695724,
    "fitness": 0.8108455051102499,
    "legacy_fitness": 0.8108455051102499,
    "target_cdf": 0.9954646088914592,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.9223079681396484,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.752830647834548,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.9223079681396484,
        "normalized": 0.9954646088914592,
        "contribution": 0.4977323044457296
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5309876203536987,
        "normalized": 0.09206841268091792,
        "contribution": 0.02301710317022948
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.691179275512695,
        "normalized": 0.9685129990310137,
        "contribution": 0.14527694985465206
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.695219039916992,
        "normalized": 0.8680429036393686,
        "contribution": 0.08680429036393687
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "5bac4f7003932304d87901db528d29dd7bbb77e123cc90fe73507f43670f2578",
    "id": "fbbd1a441a47f2a8862c9e4e53c1",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p05",
    "language": "de",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich dachte, du wärst hier, bevor die Lichter ausgehen. Warte... ich brauche einen Moment, um das gerade zu verstehen. Bleib bei mir, bis ich das richtig sagen kann.",
    "sentences": [
      "Ich dachte, du wärst hier, bevor die Lichter ausgehen.",
      "Warte... ich brauche einen Moment, um das gerade zu verstehen.",
      "Bleib bei mir, bis ich das richtig sagen kann."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Project a dominant loud command with forceful breath and precise words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Finish with cold, decisive control rather than a loose shouting fit. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] Ich dachte, du wärst hier, bevor die Lichter ausgehen.\n(Natural clear speech.) [4.3 seconds duration] Warte... ich brauche einen Moment, um das gerade zu verstehen.\n(Natural clear speech.) [3.8 seconds duration] Bleib bei mir, bis ich das richtig sagen kann.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.24500846862793,
    "genuineness": 4.250029563903809,
    "wer": 0.0,
    "identity": 0.6231287717819214,
    "voiceclap": 0.07919488847255707,
    "fitness": 0.8244817231316819,
    "legacy_fitness": 0.8244817231316819,
    "target_cdf": 0.98254389250665,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.402869701385498,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8043871398167948,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.402869701385498,
        "normalized": 0.98254389250665,
        "contribution": 0.491271946253325
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.311241626739502,
        "normalized": 0.3337114436747208,
        "contribution": 0.0834278609186802
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.104061126708984,
        "normalized": 0.9104853722081963,
        "contribution": 0.13657280583122944
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.1088666915893555,
        "normalized": 0.9311452681356018,
        "contribution": 0.0931145268135602
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d5e9dc352c0bd482dd6ac7f1c8afa140541a39c757cb6c4a55100ffbe95c20ff",
    "id": "fbe4d747aa4ccd05a3e8ce4ef3bc",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "en",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.427689552307129,
    "genuineness": 4.15588903427124,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.12237028777599335,
    "fitness": 0.8160733794685744,
    "legacy_fitness": 0.8160733794685744,
    "target_cdf": 0.9644381951325223,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.086336612701416,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7181007288544894,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.086336612701416,
        "normalized": 0.9644381951325223,
        "contribution": 0.48221909756626113
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.6259750127792358,
        "normalized": 0.11146953771947371,
        "contribution": 0.027867384429868428
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.935042381286621,
        "normalized": 0.8840175306994371,
        "contribution": 0.13260262960491556
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.2111635208129883,
        "normalized": 0.754116172534443,
        "contribution": 0.07541161725344431
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "cab89a6cc0d4d07579535dbce683c2561f2864309755c57b1a4004f07b7d8b2e",
    "id": "fbe8e662c2508f12cdb17193da1a",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 3,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 3.7416701316833496,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.054443005472421646,
    "fitness": 0.8004726301298801,
    "legacy_fitness": 0.8004726301298801,
    "target_cdf": 0.9780803940030465,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.3054747581481934,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.78676447150015,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.3054747581481934,
        "normalized": 0.9780803940030465,
        "contribution": 0.48904019700152324
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.131443977355957,
        "normalized": 0.2621540050393544,
        "contribution": 0.0655385012598386
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.353141784667969,
        "normalized": 0.9409148986309699,
        "contribution": 0.1411372347946455
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.950448513031006,
        "normalized": 0.9104853844414267,
        "contribution": 0.09104853844414268
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4d0ea72e25128384634330a51f23085c7782cc1dfe11aa0da4023269d4a6a6ca",
    "id": "fc51ced30273c2e8b36628bfc02c",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "de",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "An diesem Abend ist etwas, das ich nicht erklären kann. Nein... lass mich ausreden, bevor du antwortest. Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "sentences": [
      "An diesem Abend ist etwas, das ich nicht erklären kann.",
      "Nein... lass mich ausreden, bevor du antwortest.",
      "Du musst noch den Rest hören, den ich dir erzählen wollte."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] An diesem Abend ist etwas, das ich nicht erklären kann.\n(Natural clear speech.) [3.0 seconds duration] Nein... lass mich ausreden, bevor du antwortest.\n(Natural clear speech.) [4.7 seconds duration] Du musst noch den Rest hören, den ich dir erzählen wollte.",
    "requested_seconds": 12.0,
    "actual_seconds": 12.0,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.221180438995361,
    "genuineness": 4.18099308013916,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04607304930686951,
    "fitness": 0.8292434460504938,
    "legacy_fitness": 0.8292434460504938,
    "target_cdf": 0.9990117801396228,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.425514221191406,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7869749917590262,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.425514221191406,
        "normalized": 0.9990117801396228,
        "contribution": 0.4995058900698114
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9075450897216797,
        "normalized": 0.18551887784237608,
        "contribution": 0.04637971946059402
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.9580512046813965,
        "normalized": 0.9818588747524496,
        "contribution": 0.14727883121286744
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.170474052429199,
        "normalized": 0.9381055101575337,
        "contribution": 0.09381055101575338
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "68510c070d05dc66bfd709867cac040fb64ab6ccb199651aee9ba46c5ac1256a",
    "id": "fc7b60a9f8ecd42436d027cc1e18",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p04",
    "language": "de",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 2,
    "generation": null,
    "ce": 6.467403411865234,
    "genuineness": 4.256305694580078,
    "wer": 0.0,
    "identity": 0.629635214805603,
    "voiceclap": -0.11746662855148315,
    "fitness": 0.834826223202407,
    "legacy_fitness": 0.834826223202407,
    "target_cdf": 0.9915901345965457,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.6950535774230957,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7723952880221217,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.6950535774230957,
        "normalized": 0.9915901345965457,
        "contribution": 0.49579506729827283
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.039849042892456,
        "normalized": 0.2289868888511522,
        "contribution": 0.05724672221278805
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.059582710266113,
        "normalized": 0.9040003875053659,
        "contribution": 0.1356000581258049
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.5456018447875977,
        "normalized": 0.8375344038525582,
        "contribution": 0.08375344038525583
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d697d670f7ebd5f4c4640ff4a8f097975e094d593fd970d85a5ba2cc42351bab",
    "id": "fcd661ae21045c9563ae792da444",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1702,
    "valid": true,
    "ended": true,
    "text": "I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.",
    "sentences": [
      "I thought you would be here before the lights went out.",
      "Wait... I need a moment to understand what just happened.",
      "Stay with me until I can say this properly."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.\n(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.\n(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.",
    "requested_seconds": 11.52,
    "actual_seconds": 11.52,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.395919322967529,
    "genuineness": 4.111956596374512,
    "wer": 0.0,
    "identity": 0.5887099504470825,
    "voiceclap": -0.030525827780365944,
    "fitness": 0.7358360256435288,
    "legacy_fitness": 0.7358360256435288,
    "target_cdf": 0.8092130354408053,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.1205055713653564,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6210176968496125,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.1205055713653564,
        "normalized": 0.8092130354408053,
        "contribution": 0.4046065177204026
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.9855704307556152,
        "normalized": 0.2105000137670302,
        "contribution": 0.05262500344175755
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.1233069896698,
        "normalized": 0.6852468300972876,
        "contribution": 0.10278702451459314
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7513105869293213,
        "normalized": 0.6099915117285923,
        "contribution": 0.060999151172859235
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "04d4b97174350eac3db45631557bed47aa29eec228ca9b2a0cefb5c38609010d",
    "id": "fce654768111fa570f16b42b2540",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p09",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Let the local direction carry the intended delivery beside each sentence.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 2.5,
    "generation": null,
    "ce": 6.2767791748046875,
    "genuineness": 3.8044304847717285,
    "wer": 0.0,
    "identity": 0.6276042461395264,
    "voiceclap": 0.025410352274775505,
    "fitness": 0.8093251794528249,
    "legacy_fitness": 0.8093251794528249,
    "target_cdf": 0.9877755264344382,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.548961639404297,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7573820031279398,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.548961639404297,
        "normalized": 0.9877755264344382,
        "contribution": 0.4938877632172191
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.768456220626831,
        "normalized": 0.1457841285203917,
        "contribution": 0.03644603213009792
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.39762020111084,
        "normalized": 0.9453738037624609,
        "contribution": 0.14180607056436914
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.6160099506378174,
        "normalized": 0.852421372162536,
        "contribution": 0.08524213721625361
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "b619149c1faab91496a01e67234c5b5377c6f0fc421f9ee225c06565b89cb2cd",
    "id": "fd96bdd57fcf489ecdf0adad7d61",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "en",
    "reference": false,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "You are the first person I wanted to find tonight. I tried to prepare for this... but the words keep changing. Please listen carefully because this matters more than I expected.",
    "sentences": [
      "You are the first person I wanted to find tonight.",
      "I tried to prepare for this... but the words keep changing.",
      "Please listen carefully because this matters more than I expected."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] You are the first person I wanted to find tonight.\n(Natural clear speech.) [4.2 seconds duration] I tried to prepare for this... but the words keep changing.\n(Natural clear speech.) [3.8 seconds duration] Please listen carefully because this matters more than I expected.",
    "requested_seconds": 11.84,
    "actual_seconds": 11.84,
    "cfg": 3,
    "generation": null,
    "ce": 6.403861999511719,
    "genuineness": 3.848362922668457,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.04184526950120926,
    "fitness": 0.8190912554958534,
    "legacy_fitness": 0.8190912554958534,
    "target_cdf": 0.9972925007937494,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 4.100865364074707,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7475987471825453,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 4.100865364074707,
        "normalized": 0.9972925007937494,
        "contribution": 0.4986462503968747
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5649117231369019,
        "normalized": 0.0986881865760929,
        "contribution": 0.024672046644023227
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.5132646560668945,
        "normalized": 0.9557276577709561,
        "contribution": 0.14335914866564342
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.4223875999450684,
        "normalized": 0.8092130147600393,
        "contribution": 0.08092130147600393
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "282311348296490dbd1097f6eb7cbe36f52803b2e14a7bc6cb2713e483cc44b7",
    "id": "fdb653924cd02647bbd80481ed38",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p02",
    "language": "en",
    "reference": true,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.",
    "sentences": [
      "I remember how ordinary everything looked a moment ago.",
      "Now... I am not quite sure where I should begin.",
      "Give me a little time and I will explain the whole thing."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Project a dominant loud command with forceful breath and precise words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Finish with cold, decisive control rather than a loose shouting fit. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.\n(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.\n(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.",
    "requested_seconds": 11.92,
    "actual_seconds": 11.92,
    "cfg": 1.25,
    "generation": null,
    "ce": 6.38797664642334,
    "genuineness": 3.9927117824554443,
    "wer": 0.0,
    "identity": 0.7218949794769287,
    "voiceclap": -0.055938541889190674,
    "fitness": 0.7409307089638595,
    "legacy_fitness": 0.7409307089638595,
    "target_cdf": 0.8297366037352649,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.2016677856445312,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6417407480991405,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.2016677856445312,
        "normalized": 0.8297366037352649,
        "contribution": 0.41486830186763246
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8973677158355713,
        "normalized": 0.18240174630185035,
        "contribution": 0.04560043657546259
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.34792423248291,
        "normalized": 0.7516488247877673,
        "contribution": 0.1127473237181651
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.9801368713378906,
        "normalized": 0.6852468593788036,
        "contribution": 0.06852468593788036
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "edd1aee057dd3423ccd8b95869a6fb83291942fff5805cef66ac5719fd676a22",
    "id": "fdfbcedbfb5d10ed92c71e541652",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p08",
    "language": "en",
    "reference": true,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Carry the full intended delivery through the global direction.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 2,
    "generation": null,
    "ce": 6.483288764953613,
    "genuineness": 4.231201171875,
    "wer": 0.0,
    "identity": 0.5410799384117126,
    "voiceclap": 0.0030597615987062454,
    "fitness": 0.7698661055465288,
    "legacy_fitness": 0.7698661055465288,
    "target_cdf": 0.8629676751891268,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.347759962081909,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.6350631581397381,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.347759962081909,
        "normalized": 0.8629676751891268,
        "contribution": 0.4314838375945634
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.7955955266952515,
        "normalized": 0.15305051991141005,
        "contribution": 0.038262629977852514
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.143322467803955,
        "normalized": 0.6914624612740131,
        "contribution": 0.10371936919110196
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 2.7689125537872314,
        "normalized": 0.6159732137622032,
        "contribution": 0.061597321376220315
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "4647ba5425cfa22b2cb2849e7db9aeb50aaa4b83d472daddb3b9f5d6a8411c77",
    "id": "fe6bcbc9adfe8f5fca52927486da",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p03",
    "language": "de",
    "reference": false,
    "seed": 1701,
    "valid": true,
    "ended": true,
    "text": "Ich war mir sicher, dass ich wusste, was als Nächstes passiert. Und dann... hat sich für einen Moment alles anders angefühlt. Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "sentences": [
      "Ich war mir sicher, dass ich wusste, was als Nächstes passiert.",
      "Und dann... hat sich für einen Moment alles anders angefühlt.",
      "Sieh mich an und sag mir, dass du verstehst, was ich meine."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Project a dominant loud command with forceful breath and precise words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Finish with cold, decisive control rather than a loose shouting fit. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.7 seconds duration] Ich war mir sicher, dass ich wusste, was als Nächstes passiert.\n(Natural clear speech.) [4.3 seconds duration] Und dann... hat sich für einen Moment alles anders angefühlt.\n(Natural clear speech.) [5.1 seconds duration] Sieh mich an und sag mir, dass du verstehst, was ich meine.",
    "requested_seconds": 14.08,
    "actual_seconds": 14.08,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.38003396987915,
    "genuineness": 4.394378662109375,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.11206269264221191,
    "fitness": 0.8161405050443475,
    "legacy_fitness": 0.8161405050443475,
    "target_cdf": 0.9470810897522896,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.891547203063965,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7174788627517629,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.891547203063965,
        "normalized": 0.9470810897522896,
        "contribution": 0.4735405448761448
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.975393295288086,
        "normalized": 0.20713393848934814,
        "contribution": 0.051783484622337034
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.5480780601501465,
        "normalized": 0.8038541354521456,
        "contribution": 0.12057812031782184
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.079148292541504,
        "normalized": 0.7157671293545922,
        "contribution": 0.07157671293545922
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1e6b48c6acc9d3217d1184cba60b2a178dcdc0b6d5612378f66bd6c2eb3b65db",
    "id": "fe77c33e421d26412938cac8b77d",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p07",
    "language": "en",
    "reference": true,
    "seed": 1703,
    "valid": true,
    "ended": true,
    "text": "I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.",
    "sentences": [
      "I was certain that I knew what would happen next.",
      "And then... everything felt different for a second.",
      "Look at me and tell me you understand what I mean."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Project a dominant loud command with forceful breath and precise words. React to a silent partner standing beside you; generate only one adult voice.) [3.1 seconds duration] And then... everything felt different for a second.\n(Finish with cold, decisive control rather than a loose shouting fit. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.\n(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.\n(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.",
    "requested_seconds": 11.12,
    "actual_seconds": 11.12,
    "cfg": 1.5,
    "generation": null,
    "ce": 6.41180419921875,
    "genuineness": 3.9550557136535645,
    "wer": 0.0,
    "identity": 0.6070051193237305,
    "voiceclap": -0.04125305637717247,
    "fitness": 0.8006712954056406,
    "legacy_fitness": 0.8006712954056406,
    "target_cdf": 0.9511644047125467,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.932128429412842,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.72269941419057,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.932128429412842,
        "normalized": 0.9511644047125467,
        "contribution": 0.47558220235627335
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.8736209869384766,
        "normalized": 0.17525676081671665,
        "contribution": 0.04381419020417916
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.71264910697937,
        "normalized": 0.8413447460685429,
        "contribution": 0.12620171191028143
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.272770404815674,
        "normalized": 0.7710130971983604,
        "contribution": 0.07710130971983604
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "0f10eccb13e3524dfe61f73a4a9f9232bffec1179d809042e1c4442bca20019b",
    "id": "fec15d0b3c5e2455d646b882a545",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 6.149696350097656,
    "genuineness": 4.344170093536377,
    "wer": 0.0,
    "identity": null,
    "voiceclap": -0.011955861002206802,
    "fitness": 0.7959838606323741,
    "legacy_fitness": 0.7959838606323741,
    "target_cdf": 0.9224687292985007,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 2.6886415481567383,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7792749382652242,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 2.6886415481567383,
        "normalized": 0.9224687292985007,
        "contribution": 0.46123436464925033
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.491039514541626,
        "normalized": 0.4119015661257916,
        "contribution": 0.1029753915314479
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 3.694857597351074,
        "normalized": 0.8375344060053678,
        "contribution": 0.12563016090080517
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.8448362350463867,
        "normalized": 0.8943502118372078,
        "contribution": 0.08943502118372078
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "1436034a83e87e9981d263b91a8134fb6c64edc3fbde67dda851cbba844bffc1",
    "id": "ff6dd7718fa20402a75fef5a3534",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p06",
    "language": "de",
    "reference": false,
    "seed": 1705,
    "valid": true,
    "ended": true,
    "text": "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat. Jetzt... weiß ich nicht so recht, wo ich anfangen soll. Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "sentences": [
      "Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.",
      "Jetzt... weiß ich nicht so recht, wo ich anfangen soll.",
      "Gib mir ein wenig Zeit, dann erkläre ich dir alles."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Project a dominant loud command with forceful breath and precise words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Finish with cold, decisive control rather than a loose shouting fit. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [4.3 seconds duration] Ich weiß noch, wie gewöhnlich gerade eben alles ausgesehen hat.\n(Natural clear speech.) [4.3 seconds duration] Jetzt... weiß ich nicht so recht, wo ich anfangen soll.\n(Natural clear speech.) [4.3 seconds duration] Gib mir ein wenig Zeit, dann erkläre ich dir alles.",
    "requested_seconds": 12.88,
    "actual_seconds": 12.88,
    "cfg": 3,
    "generation": null,
    "ce": 6.038498878479004,
    "genuineness": 4.256305694580078,
    "wer": 0.1,
    "identity": null,
    "voiceclap": -0.010405930690467358,
    "fitness": 0.7392708498304891,
    "legacy_fitness": 0.7392708498304891,
    "target_cdf": 0.9862070255954634,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.5002641677856445,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.8128191557535898,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.5002641677856445,
        "normalized": 0.9862070255954634,
        "contribution": 0.4931035127977317
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 2.2976720333099365,
        "normalized": 0.3280469413306799,
        "contribution": 0.08201173533266998
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.602221965789795,
        "normalized": 0.9625641895690078,
        "contribution": 0.14438462843535116
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 4.126468658447266,
        "normalized": 0.9331927918783697,
        "contribution": 0.09331927918783697
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  },
  {
    "metadata_schema": 3,
    "source_metadata_sha256": "d32710e9212691b09666c60ff5cef6b9f5bfa0030a6d47b81a890c7a19f48a84",
    "id": "ffaf902b24c7de0883f8d5321ac5",
    "study": "full-v2-gen0",
    "model": "original15",
    "mission": "drill",
    "candidate": "g0-p01",
    "language": "en",
    "reference": false,
    "seed": 1704,
    "valid": true,
    "ended": true,
    "text": "There is something about this evening that I cannot explain. No... let me finish before you try to answer. You have to hear the rest of what I came to tell you.",
    "sentences": [
      "There is something about this evening that I cannot explain.",
      "No... let me finish before you try to answer.",
      "You have to hear the rest of what I came to tell you."
    ],
    "prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional drill instructor begins with controlled, grounded authority. Project a dominant loud command with forceful breath and precise words. Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.\nSCRIPT:\n(A fictional drill instructor begins with controlled, grounded authority. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Project a dominant loud command with forceful breath and precise words. Make the intended feeling unmistakably audible while preserving every word.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Finish with cold, decisive control rather than a loose shouting fit. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "neutral_prompt": "GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.\nSCRIPT:\n(Natural clear speech.) [3.8 seconds duration] There is something about this evening that I cannot explain.\n(Natural clear speech.) [3.5 seconds duration] No... let me finish before you try to answer.\n(Natural clear speech.) [5.0 seconds duration] You have to hear the rest of what I came to tell you.",
    "requested_seconds": 12.32,
    "actual_seconds": 12.32,
    "cfg": 1,
    "generation": null,
    "ce": 6.443574905395508,
    "genuineness": 3.9173994064331055,
    "wer": 0.0,
    "identity": null,
    "voiceclap": 0.03250328078866005,
    "fitness": 0.8193379829910927,
    "legacy_fitness": 0.8193379829910927,
    "target_cdf": 0.9900472701763179,
    "target_head": "vn_S_AUTH_reg",
    "raw_target": 3.630124092102051,
    "bursts": [],
    "event_matches": [],
    "family_hit_all": null,
    "nominal_hit_all": null,
    "lexical_proxy": false,
    "cached_historical": false,
    "scoring_waveform": "Complete pre-MP3 generated waveform; browser audio is its mono128kbps MP3",
    "timing_evidence": "Nominal requested windows, not forced-aligned word timestamps",
    "reward_contract": {
      "kind": "emotion-or-style",
      "target_head": "vn_S_AUTH_reg",
      "target_normalization": {
        "mean": 1.211488127708435,
        "std": 1.0388771295547485
      },
      "weights": {
        "target": 0.5,
        "content_enjoyment": 0.25,
        "genuineness": 0.25
      },
      "normalizers": {
        "content_enjoyment": "clip(CE / 10, 0, 1)",
        "genuineness": "clip(genuineness / 6, 0, 1)",
        "target": "fixed training-distribution Gaussian CDF",
        "blend": "fixed training-distribution Gaussian CDF"
      },
      "wer_factor": "1 - min(1, raw_WER)",
      "family_factor": null,
      "quality_gates": {
        "per_take_wer_max": 0.15,
        "candidate_mean_wer_max": 0.1,
        "valid_terminated_fraction_min": 0.9,
        "reference_identity_delta_min": -0.05,
        "reference_pairs": 50,
        "all_family_fraction_min": null
      },
      "scientific_caveat": "Search ranking, not the RL reward; predicted proxies, not human preference. Raw scores and failed takes are retained."
    },
    "steering_v2": 0.7386893629752725,
    "steering_components": [
      {
        "head": "vn_S_AUTH_reg",
        "weight": 0.5,
        "direction": "high",
        "mean": 1.211488127708435,
        "std": 1.0388771295547485,
        "raw": 3.630124092102051,
        "normalized": 0.9900472701763179,
        "contribution": 0.49502363508815894
      },
      {
        "head": "vn_VOLT_reg",
        "weight": 0.25,
        "direction": "high",
        "mean": 2.6844067573547363,
        "std": 0.8684567809104919,
        "raw": 1.5513420104980469,
        "normalized": 0.09599963185428989,
        "contribution": 0.023999907963572473
      },
      {
        "head": "vn_STNC_reg",
        "weight": 0.15,
        "direction": "high",
        "mean": 2.57399582862854,
        "std": 1.13865327835083,
        "raw": 4.326454162597656,
        "normalized": 0.9381054562342774,
        "contribution": 0.1407158184351416
      },
      {
        "head": "vn_AROU_reg",
        "weight": 0.1,
        "direction": "high",
        "mean": 2.4366743564605713,
        "std": 1.126529574394226,
        "raw": 3.3431787490844727,
        "normalized": 0.7895000148839948,
        "contribution": 0.07895000148839948
      }
    ],
    "steering_contract": {
      "version": "ears-steering-v2",
      "targets": [
        {
          "head": "vn_S_AUTH_reg",
          "weight": 0.5,
          "direction": "high",
          "mean": 1.211488127708435,
          "std": 1.0388771295547485
        },
        {
          "head": "vn_VOLT_reg",
          "weight": 0.25,
          "direction": "high",
          "mean": 2.6844067573547363,
          "std": 0.8684567809104919
        },
        {
          "head": "vn_STNC_reg",
          "weight": 0.15,
          "direction": "high",
          "mean": 2.57399582862854,
          "std": 1.13865327835083
        },
        {
          "head": "vn_AROU_reg",
          "weight": 0.1,
          "direction": "high",
          "mean": 2.4366743564605713,
          "std": 1.126529574394226
        }
      ],
      "formula": "sum(weight * Phi((raw_EARS_head - frozen_mean) / frozen_std)); low targets use 1 - Phi",
      "pure_condition": "100% of its named emotion or style head, with no auxiliary reward",
      "content_enjoyment_weight": 0,
      "genuineness_weight": 0,
      "wer_reward_weight": 0,
      "identity_reward_weight": 0,
      "normalization": "Fixed training-distribution Gaussian CDF; not an emotion probability; raw head means remain visible",
      "quality": "Original WER, termination, paired identity and burst-qualification gates remain separate",
      "scope": "Whole-clip emotion/style evidence; does not prove timed transitions, vocal-burst realization or naturalness",
      "historical": "CPU re-score of existing frozen raw EARS predictions; original fitness retained. This is not a new generation or an independent validation."
    },
    "steering_missing_raw": false,
    "requested_events_detector_supported": true,
    "burst_evaluation_status": null,
    "audio_url": null,
    "audio_sha256": null,
    "archive": "https://huggingface.co/datasets/laion/Humaneness-Voice-Large-Prompting-Audio/tree/main/packed"
  }
]